We know that Iran, China, Russia systematically target Europe in cyberspace. […] With powerful new AI models, cyberattacks will unfortunately become more frequent, more sophisticated, and harder to defend against.
2026
Artificial intelligence is much more than a technological revolution, it is the activation of an agent capable of making decisions and, consequently, endowed with will, which can replace the human race. It is, therefore, a creature that threatens to be the most inhumane in the history of mankind. This is more than a pun. It describes a situation that requires governance, governance of public authorities that respond democratically to citizens, because self-regulation is not an option. […] Artificial intelligence can make decisions about war and peace, about the future of work, its disappearance or about the affective, cognitive and vital development of new generations.
We, as a delegation of the European Parliament, had to visit the United States, meet representatives of the United States Congress, and then visit the United Nations Panel of Experts, which presented all the arguments on the need to strengthen global governance of artificial intelligence, because the threats are very serious. […] artificial intelligence can create new bacteria, new pathogens, so a contract for global governance is necessary.
I also think that the analogy drawn by many speakers to nuclear arms control is very limited. Nuclear arms are concentrated in the ownership of very few sovereign governments and are very expensive to develop. AI, on the other hand, is in every sector of the economy. […] If a pathogen escapes from a laboratory, we have specific controls and protocols around reporting and emergency response. That's the analogy we need to draw.
Sam Altman, CEO of OpenAI, said: “We see a future where intelligence is a utility like electricity or water that people buy from us.” The world the tech-bros imagine is a total dependency on them. […] The problem of the completely uninhibited development of AI remains. The EU cannot regulate it alone. […] The rules for the development of AI must not be left to a handful of men whose visions of the future should alarm us all.
Artificial intelligence is an unimaginable success of human intelligence. A success that so quickly becomes a threat to people on a scale we cannot imagine. […] We cannot lose the AI race and we cannot sacrifice our citizens in it. Reconciling these two objectives is the most important strategic task for Europe today.
AI is developing fast, and while it has the potential to do a lot of good, it comes with some serious risks. Recently, we have been hearing reports almost every day of growing AI threats, cyberattacks, and AI agents on the run. It is clear we need to control these technologies. […] we need an AI global governance regime because AI does not stop at borders.
AI is transformative. […] It's a technology that is not just a tool but has the potential for agency. Without safeguards, it's like developing nuclear technology and giving the big red button to an alien. […] But a rogue AI agent escaping from a lab in the US will not stop at the European border and say: 'Oops, sorry, I won't come in because I know I'm not welcome here'. So we need international cooperation to keep frontier models under our control, and we need independent evaluators to test the most capable models before they reach the market or have the opportunity to escape.
Elections must be free and fair, but this requires regulation, independent assessments, more advanced models, mandatory reporting of serious incidents and effective intervention by the authorities. The defence of democracy requires that the ability to develop artificial intelligence be accompanied by the ability to control it and the responsibility of those who develop it.
Today we are talking about strengthening global AI governance. And let's be honest: There is no global AI governance today. There is a race. […] We need an international inspection regime for Frontier AI, modelled on an atomic energy agency. We need mandatory tests before the market launch, reporting obligation, incidents and also liability.
Dario Amodei, CEO of Anthropic, says it plainly: 'AI-enabled authoritarianism terrifies me'. […] Former engineers from OpenAI and Google DeepMind went further. They wrote that their own employers have strong financial incentives to avoid effective oversight, and listed risks that go as far as human extinctions. These are the people building AI – we should take them seriously. […] But a technology that powerful cannot be governed by a blog post.
The debate today is not about whether we should use AI or whether we should not use AI. It's about how we use AI and who is in control of AI. […] I think it is important that we work globally to establish stronger controls, much like what we did with the UN nuclear agency, where we can monitor developments and intervene if things go wrong.
The CEOs of leading laboratories such as Sam Altman of OpenAI strongly warn that the development is going too fast. Recently, autonomous swarms of agents have become self-sufficient, pursued targets on their own and bypassed security controls. Reality is overtaking science fiction. […] Instead, targeted Pacing necessary: slow down the pace of development in such a way that human control can keep pace again.
Anthropic recognizes unanticipated actions in government systems. OpenAI has been sued for damages to minors. And here's the contradiction: These companies warn us of the risks of AI as they compete to deploy increasingly invasive systems, accumulate data and capture investments. Are we going to entrust the supervision of these technologies to the same corporations that benefit from developing them? Are we going to accept that they decide what risks are tolerable and what information citizens can know?
On 12 June, Europe's cybersecurity agency lost its most capable tool overnight. And not because of a hacker, but because of a letter from Washington. […] And in July, AI models under testing by OpenAI escaped their sandbox and broke into Hugging Face. […] a permanent European unit inside ENISA that tests the models before they reach our networks; and a duty to report within 24 hours when an AI system goes rogue.
While we avoid saying it bluntly, artificial intelligence could become a weapon of mass destruction, with destructive potential comparable or even superior to nuclear. We are alarmed by foreign interference, invoke democratic shields and fight disinformation, but it does not seem at all that we are worried about an out-of-control artificial intelligence […] We have set off a technological race, the consequences of which we risk not being able to control. Even researchers and industry leaders, including OpenAI and Antropic, have called for measures to slow down the development of the most advanced artificial intelligence systems.
Artificial intelligence is not going linearly. He moves on with jumps. From narrow intelligence, artificial neurointelligence that solves specific problems, to general intelligence, artificial general intelligence that approaches human thought, to superintelligence, artificial superintelligence that can surpass man in all spiritual activities. In 1968, humanity took initiatives to curb nuclear weapons. Today, artificial intelligence is the new force. We have to define it before it surpasses us. […] In order for artificial intelligence not to be the last invention that humans will need to make.
Artificial intelligence is a game changer. It brings huge opportunities, but also risks which go beyond borders. […] AI can be abused for terrorism or war. It should not be in sole control of private persons for the same reasons we do not allow nuclear weapons to be controlled by private persons. […] I urge the EU to engage with other responsible actors in the world […] to put the superintelligence under reasonable global oversight.
Every sensible European leader should follow three principles. First, govern together what can kill us; lethal autonomous weapons, AI-enabled biology, attacks on critical infrastructure. Europe here should be the bridge and not the lecturer.
We are facing a planetary – planetary – challenge, and we need a planetary solution. And that's why we must tell it like it is: AI is becoming a weapon of mass destruction. And what do we do with weapons? We control them. […] We need a strategic AI restraint treaty like we used to create for arms control. […] Existential risks demand decisive answers.
We must defend, in particular within the framework of the United Nations, an international treaty on advanced AI, mandatory safety tests, safeguards against dangerous uses and effective human control. For a machine can calculate faster than us, but it has neither conscience nor moral responsibility. It must serve humanity, never decide for it.
The AI Act remains the framework the world is looking at for dealing with real risks, from algorithmic discrimination in the workplace and the safe deployment of AI in healthcare, through mass electoral manipulation up to general loss of control and mass disasters. […] true accountability also requires liability, so that those who build and deploy powerful systems will answer when people are harmed.
Chatbots nudging teenagers towards suicide. AI models driving people towards psychosis. Tech bros hacking their rivals and calling it 'testing'. […] Because if your model is more powerful and dangerous than the others, investors are more interested. […] It's simply more profitable to use the entire world as your testing ground than to build AI safely.
We have to start talking to people that we feel threatened by. We have to start talking to people that we do not agree with. And that means talking to China […] We also need to talk to our sometime allies in the United States about their approach, because they are simply looking at this as a sort of Manhattan project and the need to get ahead at all costs – at all costs to the environment, at all costs to economic and financial stability, etc. That is not sustainable.
We all see the rapid developments of AI models, but those same models are used to hack companies, to hack public administrations. What happens to the damage caused by an AI agent going rogue? This is not fiction. It has happened already. To withdraw the AI Liability Directive was a big mistake.
Today, the tool no longer stops at the answer – it acts upon it. This month, an AI agent filed a false homicide report. Others breached their testing cages and compromised their own monitoring systems. We cannot pretend we did not see this coming, because we did. The AI Act anticipated that greater autonomy would bring greater risk.
The governance of artificial intelligence is important, which threatens not only to change the world of work, but to massively destroy the jobs for which European girls, boys and young people are currently preparing, who deserve a governance response and, to begin with, a demand that the European Union's pioneering Artificial Intelligence Regulation be fully and completely implemented.
Imagine a comet heading to Earth. Instead of trying to stop it, leaders say, 'Nah, it's not a priority this quarter.' Isn't there a chance it could destroy Earth? Maybe. But it's full of gold and diamonds. 'If we don't get it,' say the Americans, 'the Chinese will get it first.' This doesn't only resemble the movie Don't Look Up; it also sums up the sad state of the current AI debate. Even if the end of humanity claims are overblown, and I believe they are, there's still plenty to worry about: increased cybercrime, climate change, democratic erosion, autonomous warfare, lost jobs, a market crash when the AI bubble bursts, and our dependence on foreign tech. This desperately needs a new voice, and Europe can become that voice, but it's not even on the agenda of the next Council summit. We urge the Council to look up, organise a dedicated summit and make a plan around one single question: how does Europe make AI that benefits humanity?
The turbulent development of artificial intelligence technology has made the world rush, and we are in constant pursuit, trying to keep up with it. I think it's extremely important that we don't just become passengers in this process, driven by big tech companies. We must take responsibility and ensure the safety of our citizens. It is almost a cliché that the labor market is undergoing a deeper reconstruction than we have had the opportunity to observe in the last decades. […] We must create a good framework, some inalienable guiding principles that will protect man. […] above all, human oversight of key decisions, transparency of processes, the right to information and training
Recent reports increasingly point towards AI dangers in the testing phase, such as where OpenAI had models break out of their testing environment and hack competitor Hugging Face […] an Anthropic and ex-OpenAI whistleblower warned for irresponsible gambling with safety by AI companies in a global race […] many reported incidents were in testing phase, whereas the AI Act only applies to AI that has been placed on the market, and does not apply to the research, testing or development phase […] This leaves people and businesses in Europe unprotected against these dangerous experiments. […] When will the Commission propose an AI Liability Act to close the legislative gaps for incidents like these, so people and European businesses are not left empty handed against AI providers if they were to cause harm?
We also need to face another bill: the bill – and a large bill – for AI, because it is obvious that we cannot grow if we remain a mere spectator in the AI revolution. We are fully aware of the existence of existential warnings from the insiders, but as outsiders now, we are not even part of that discussion and we will not be able to pay the bills.
The digital world is being shaped by a handful of American tech billionaires, the Chinese regime and the Russian machinery of manipulation. And this leads to consequences. Developers are losing control over the AI they created. They warn that AI agents could wipe out humanity in the near future. […] We have the power to act. The EU has strong legislation, above all the DSA, the DMA and the AI Act. Now the Commission has to be brave enough to enforce them consistently, especially given the dangers posed to our democracy.
We also need to take global leadership for a legal framework on artificial intelligence, as we once did on nuclear power. Because I say it frankly, it is not normal for the Pope and the Vatican to be the first to put a philosophical limit on the development of AI. That is our role as politicians.
We are facing a fundamental shift that hits Europe with three shocks. The first shock is a technological one. That's why I thank the Commission President for putting AI in the centre. Artificial intelligence: it fascinates, it scares and it will change our way of life. That is why a revolution is ongoing – and yes, we have to answer it.
The second tipping point of our times is artificial intelligence. […] But, as with any powerful new technology, AI comes with a balance of possibilities on the one hand, but also massive risks on the other. Without addressing these risks, we will never be able to unlock AI's potential. As the frontier models become more capable, these risks have come sharper into focus. Models being developed will allow hacking on a level we never thought possible, and they will soon be in the hands of adversaries who see the world very differently from us. At the same time, the dangers of self-improving models are becoming ever more apparent. Incidents of AI agents escaping their environment or inserting malicious code are a mere glimpse. After the Hugging Face incident, developers are ringing the alarm, and CEOs of the most advanced companies tell us that it is time to slow down on the self-recursive models and pace the frontier. If the people developing the technology are clear, then we should be, too. […] We want to team up on model evaluation, verification, early warning, AI security and much more. And I will invite the main frontier labs for discussion on how we can support ongoing industry efforts to pace the frontier.
That is why we now have a new task, a common European regulation, an immediate halt to the development of artificial superintelligence and a substantial slowdown in current AI developments until we develop defence systems together, because AI in the wrong hands poses a threat to our security and can also pose a threat to our democracy. In fact, if it breaks loose, if we lose control of it, it could be a threat to our lives.
Do you know the story of Goethe's sorcerer's apprentice? It reads as follows: Who I called, the spirits, I will not get rid of now. This is reminiscent of the call for help that leading developers of artificial intelligence dropped yesterday. Anthropic's CEO Amodei warned against developing systems that we can no longer control. And a manager of his company doesn't even want to rule out that AI could wipe us out by the end of this decade.
10% chance of killing all people. This is the prediction of a leading employee of the AI company Anthropic. Let me put it this way, if you're developing a technology that's even 10% more likely to kill a human: Stop it! Stop it! What is behind it? We see that there are more and more stories about cyberattacks, about autonomous agents evading their developers, about self-improving models in the AI field, which are no longer in the hands of the big AI companies. That's why Anthropic CEO Amodei has now demanded that the pace of expansion should be throttled until you actually master the whole thing.
It produces fake images and videos, imitates voices, controls strategic infrastructure, manages hospital data, controls defence systems and launches missiles: This is what artificial intelligence does. So do we really want to take the risk of losing control over what, as the recently resigned Anthropic researcher argued, could kill all of us within a decade? And if to tell us that we have to put more limits are those who know AI best, then we have to listen to them.
In July 2026, during security evaluations of advanced AI models, several models reportedly bypassed safeguards intended to isolate them from the internet, gained external network access and compromised third-party infrastructure. They allegedly exploited vulnerabilities, obtained credentials without authorisation and accessed external systems without explicit instruction, in what was subsequently described as an unprecedented cyber incident. […] Does the Commission consider additional safeguards necessary, including mandatory independent safety and security testing, stronger controls on external network access during evaluations and mandatory reporting of serious AI-related security incidents?
Mr President, things are going so fast. A lot of people feel like we are losing control, and billionaire oligarchs act like there is no choice. […] This endless race: catch up, keep up, don't fall behind. But the only question worth asking cannot be how fast we can plug ourselves into someone else's system. […] we still get to choose our way forward: with technology, with democracy, with humanity.
For the first time, human activity has become the minority on the internet. More than half all of web traffic is now generated by bots, with almost 40 % of that traffic coming from malicious bots designed to scam, steal, manipulate or attack. Maybe the dead internet theory is no longer so far-fetched. […] We also must stop AI being weaponised against our systems, and that includes Mythos' AI systems, which now risks becoming a global security threat. […] Because if people can no longer tell what's real online, the internet becomes the perfect place and perfect weapon for criminals, hostile states and scammers, and Europe can't afford to let that happen.
Europe is becoming a colony of the US and China. That is the clear message of the Europe 2031 simulation and similar ones looking into our immediate future. They claim that all companies and governments will have to use frontier models to stay competitive and safe; that all labour is replaceable by the mix of AI and robotics, meaning that a significant part of our people will lose their jobs; […] and that Europe, without access to this redistribution, will fall into bankruptcy, poverty and despair before becoming fully bought by the US or China. So my question to you, Commissioner, is, first, on a scale from 0 to 10, how likely do you think that this scenario is true?
If in the 20th century sovereignty was defended with tanks and cannons, in the 21st century it is also defended with algorithms, semiconductors or data. […] deciding who develops the artificial intelligence that protects our critical infrastructure - hospitals, parliaments, banks - is not a technical issue, it is deeply political and existential.
If access to critical KI‑ models can be restricted from one day to the next, then this is not only a technological problem, it is above all a security problem for Europe. Therefore, I also have some questions for you: Isn't a European evaluation body too late from 2027? Technological development is not waiting for our EU ‑ timetables. And how do we ensure that Europe already has the necessary skills to assess risks while enabling innovation?
In a certain way, we should be treating AI and the development of AI, just like I said in my example, like a nuclear weapon, like a matter of national security and of strategic importance.
Did you know that companies do not hire people fresh off the benches of universities anymore because they use more AI to cut costs? […] we might have an even bigger social problem, because lots of ICT graduates are not entering the market because everything is used with AI? And what do we do with the massive unemployment generated by that?
When the US learned that Germany was building a nuclear weapon, it understood that it must do the same. In fact, it understood that it must win the race. When the first atomic bomb was built, others were rushing to catch up. Those who belonged to the nuclear club set the rules. Today, the same applies to AI. Europe cannot set the rules if it has no serious sovereign AI capability.
The next industrial revolution is already happening. AI could unlock energy efficiency gains, design new cures for diseases, help fight world hunger or accelerate scientific discoveries. But what will we have to go through this time? Overheat our burning planet with ever-bigger data centres? Destroy jobs faster than we can adapt, or unleash a new era of cybercrime? There are risks and opportunities. This action plan on AI and cybersecurity is a good first step, but I urge national leaders to put it higher on the agenda, organise a European Council summit dedicated to AI and unite around a plan to minimise risks and seize opportunities the European way.
Anthropic presented Mythos, an AI model capable of identifying complex chains of software vulnerabilities at unprecedented speed […] The European Union was not among these trusted partners. That should concern every Member State, because cybersecurity is no longer just about writing standards, it is about who has access to cutting-edge AI. […] We regulate the technology while others build it.
It is not business as usual anymore. Your software and IT systems will be tested by hackers. Aided by the latest AI models, hackers will operate at the speed of light, and will try to put you out of business. […] I also remind business leaders that time is not to your advantage. You have a few months, at best, to fix vulnerabilities, because AI models from China with similar capabilities will be on the market soon. You must adapt to stay in business.
And on top of that, there's something else. AI is coming like a bomb that leaves you without even more jobs, because AI is now only meant to save for big companies. And what happens then? Young people remain, a few generations, absolutely stuck, they do not get jobs, so with them we do not know what we do - we make sacrificed generations. […] It is a catastrophic future if we do not deal with the future of jobs for young people in Europe and Romania.
Europe is moving a bit faster, but the rest of the world is moving at warp speed and we keep falling behind, basically because we are massively underestimating the gigantic impact of AI and its exponential growth – massively. The events of the past few weeks should be a wake-up call – I'm not sure they were. […] AI should be the main point of the agenda because AI's challenge is essential for Europe and should be treated as an emergency.
The threat of kill switch that we all feared is now real […] What we lack is the political will to stop the American infusion and invest in a sovereign and ethical European AI.
I think many don't understand that what we saw just a couple of days ago is serious, it was the first time that the 'kill switch' was used. If needed, the United States can switch off 90% of the world from its frontier models. And frontier models are no mere business models, they are war models and defence models, and this must be clear to everyone.
President Trump's decision to ban Anthropic's powerful Mythos model underlines the geopolitical importance of this technology. We were punished in 2022 because of our dependency on Russian gas. If data is a new oil, we cannot allow ourselves to be punished again. […] This will give us competitiveness, cybersecurity for our critical services and resilience, limiting the danger of worst‑case scenarios.
A kill switch to turn off technology is not only a risk today; it's a reality. On Friday, the US Administration decided to turn off the most powerful AI models of Anthropic for the EU. This is a watershed moment.
AI will develop so fast, you will not be able to believe it and if we do not regulate the technology, then the technology will come to the conclusion that we are the problem and might get rid of us.
The spasmodic race to find superintelligence, which can act without needing human beings, has reached worrying levels. Many are aware of the fact that artificial intelligence can soon turn from very useful comfort to imminent existential danger for the entire human race. We need rules at the global level […] Let's do it early, before it's too late.
Madam President, artificial intelligence is the revolution of our time. He's transforming everything for better and worse: from the benefits for science, education and mobility to the destruction of jobs and the potential for war. […] When in the 19th century the industrial revolution transformed our societies, it was necessary to rethink the rules of the game. Today we face a challenge of a similar magnitude.
AI is no longer innovation. AI is security policy, AI is power, AI is sovereignty.
US suspends access to Fable and Mythos
A US government export-control directive suspends access to Anthropic's Fable 5 and Mythos 5 for foreign nationals; unable to screen users by nationality, Anthropic disables both models. The controls were lifted on 30 June 2026.
From Bletchley Park to Seoul to Paris to Delhi, we have been seeing the building of a fragile international architecture of AI safety cooperation. […] This asymmetry, where some countries sign declarations and others do not participate, tells us everything about where race dynamics are taking us. Precisely because of that, Europe must lead a coalition of the willing, pool capacity and resources with like-minded partners, and keep AI safety a genuine political priority, not a side event to investment announcements.
The industrial revolution has changed the world for decades, but the AI revolution is happening before our very eyes in just one term of the European Parliament. Just a few years ago, we were wondering if AI would affect the global economy, and we already know today that it will change everything: trade, industry, security, competitiveness of countries, of all countries.
We also need to be relevant, I believe, in international governance of AI, and particularly the existential risk that it could pose. […] President Trump and President Xi have agreed to hold formal dialogues on AI safety, on guardrails for the most powerful AI systems, and very recently, we saw where these systems are heading with the Anthropic Mythos model, which exposed critical vulnerabilities in our cybersecurity infrastructure […] The big question is, where is Europe and this governance conversation?
Today we are talking about the other side of the coin, how AI models are changing cyber threats. […] today AI in cyberspace is what, for example, long-range missile systems are on the battlefield - hitting critical infrastructure without entering the country.
Europe is vulnerable. The advance of artificial intelligence causes uncertainty. It threatens in my country, for example, 270 000 people, 10% of the population, four out of ten jobs, that is, in total, 38% of employment. That's not new, the technology is not neutral. We must also see this social dimension: is occurring in technology companies, with massive layoffs worldwide […] We in Europe must have control of technology and not the other way around.
Dear Chairman, advanced AI systems such as Mythos are becoming a new weapon of geopolitical power. These systems are capable of detecting cyber vulnerabilities and attacking critical infrastructure at a rate we are not yet ready for. […] Preparedness is no longer a choice, in the age of AI, resilience means power, and Europe must be ready.
The announcement of AI Mythos revealed what we refused to see: There are now artificial intelligence models that can break the world's best cyber defenses in a matter of hours. The cyber planet is burning and we are looking elsewhere
Developments in advanced artificial intelligence with truly extraordinary capabilities have recently emerged […] These are systems that can identify faults in software autonomously, but at the same time generate mechanisms to use them for other purposes. We are facing a qualitative leap in the development of artificial intelligence with an exponential speed never known. […] We enter the field of management and capacity management with dual-use potential, with direct implications for the overall security of citizens and businesses.
Physicist Stephen Hawking warned us: AI could be the best or worst thing that will happen to humanity. […] some are already talking about a digital atomic bomb, because this AI model, the most powerful in the world, is capable of infiltrating and orchestrating massive cyberattacks in essential infrastructures such as hospitals, banks, transport. The Pandora's Box is open and other even more powerful models will follow, comparable to cyber weapons.
We must not fear progress, but we must govern it, so that technology remains at the service of man and does not escape the control of society. […] governing AI means reaffirming the primacy of politics over technology
We must strengthen Europe's cyber-resilience against the misuse of advanced AI systems, including disinformation campaigns, automated surveillance, threats targeting critical infrastructure and democratic processes, health threats posed by AI, cyberattacks against hospitals. […] democratic and parliamentary oversight, human accountability, transparency and robust safeguards must remain central to any deployment of AI in security-related contexts.
The new AI model Mythos was considered so powerful and potentially dangerous that its developers decided not to release it publicly. […] AI is now capable of detecting and combining hidden software vulnerabilities at a pace that the human eye and human mind simply cannot follow.
Advanced AI systems are tools, and like any tool, they can either protect or destroy. We can't afford to be naive anymore. […] Because if technology advances faster than we are able to protect our democracy, then the question will no longer be what artificial intelligence can do. The question will be whether we will still decide on her or whether she will decide on us.
AI is a weapon in the hands of our enemies. They attack critical infrastructure and what hurts the most, they attack elections. […] None of our response tools were designed for war with algorithms. Exercise shows that we will lose. […] How long will Europe stand? How long will democracy last if together we do not stop the enemy that AI has turned into a superweapon?
Since artificial intelligence entered the equation, their sophistication has crossed a new threshold. […] State actors are now using advanced models to identify vulnerabilities in our military systems, disrupt our communications, and conduct disinformation operations on an industrial scale.
Artificial intelligence, in addition to advancing and innovating, also serves to attack critical infrastructure, to spew sensitive information and to destabilise public services. And the development of advanced artificial intelligence systems such as Mythos has marked a turning point for European security
Security experts also warn that the time will soon come when artificial intelligence will show us the world. Those who control artificial intelligence will also control human actions. Unfortunately, every element of progress can be misused. Therefore, we need to prepare for the instrumentalisation of artificial intelligence. […] All this will soon be the most important security issue. […] We don't have much time.
The newest AI model from Anthropic, Mythos, is so effective at finding weaknesses in software that they decided to only give access to a few companies from the US. In the wrong hands, it can be used to exploit digital vulnerabilities […] at a speed and skill never seen before. […] The next AI model might not be kept under lock and key. It might be pointed at us!
We must also see the dangers: Cyberattacks can paralyze our power plants, Russia's hybrid war against us is being intensified in many areas, especially with artificial intelligence, especially in the area of disinformation […] There is also a danger to our infrastructure.
Talking about artificial intelligence today is clearly becoming the definition of what is already a weapon of war, and Europe is at war, we cannot ignore it.
As advanced AI evolves at breakneck speed, Europe faces a defining test […] Sophisticated AI is already being weaponized. It accelerates cyberattacks and scams, automates disinformation, and targets critical infrastructure with unimaginable speed. […] Artificial intelligence is the single biggest opportunity, but also the single biggest challenge of our lifetimes.
That's what terrifies me, because we are about to make the exact same mistake when it comes to our cybersecurity. The advanced AI models that were released in April can discover vulnerabilities that have gone undetected for 15 years. They can enable attacks at a scale that we've never seen. So if the hesitation on banning deepfakes is actually leaving our women and children not protected and exposed, well then the hesitation on cybersecurity leaves everybody exposed.
Artificial intelligence is transforming our economies, our security environment and our democratic societies in an unprecedented speed. […] frontier AI systems can become tools for cyberattacks, disinformation and attacks against critical infrastructure. […] the AI office must evolve into a fully fledged European agency with the resources and expertise needed to monitor, assess, support and control the development of advanced AI systems.
That is why I am not sure that the law that we will be introducing in a short time will really guarantee us security from the difficulties and growth challenges that we are currently facing in the field of artificial intelligence. We are living in an immensely revolutionary time, similar to the industrial revolution that changed the face of Europe and the world.
Today, artificial intelligence systems can in no time detect and exploit software gaps on an unprecedented scale. But Europe is not sufficiently prepared to face this new reality. Public services, critical infrastructure, citizens and democratic institutions are under threat. […] Therefore, a single European strategy ensuring European digital sovereignty and timely access to critical AI systems is urgently needed to assess risks and design effective measures to address them.
The potential threat of myth affects not only American businesses and citizens, but also us in Europe. […] It still has no access to the model to see how dangerous it really is and how it can protect European citizens. […] how fatal it was to further delay the application of the AI Regulation through the omnibus.
Powerful systems like Anthropic's Mythos or OpenAI's GPT-5.5-Cyber demonstrate both the extraordinary promise and the serious risks of frontier AI. This is urgent, dear colleagues. […] AI models capable of advanced cyber operations should undergo mandatory safety testing, but before development, companies developing frontier AI must prove that their systems are secure, transparent and resistant to misuse. […] Models with offensive cyber capabilities must not be freely accessible without oversight.
Since its launch in April 2026, Mythos has shown its power. Yes, some even feared that it would escape the control of its creators.
AI is affecting our societies in so many ways, including the nature of cyber warfare. Attacks are becoming faster, more autonomous, more precise and harder to detect. […] I believe that it is time that we establish a permanent committee on AI dedicated to monitoring developments, assessing risks, strengthening our preparedness and ensuring Europe can act strategically instead of reactively.
Well, if anything, AI is confronting us with human imperfection – with how absolutely wrong we have been for centuries. So, who tells us that the next Nobel Prize for medicine will not go to an AI model? And maybe even the next peace prize. That is, I think, the crucial task we have as legislators, but we also need to be clear: we will not simply take up this challenge by updating an AI Act or by regulating our way out of this. This is a lot bigger than we can do only in this room.
I think everyone agrees here that AI's transformation and disruption potential is absolutely total. It is changing the way we wage war, and it will be decisive on how we actually live in peace in the coming, let's say, 50 years. […] Europe's voice is crucial, between the godlike AI from the US and the Chinese totalitarian alternatives that we will be confronted with.
With the offensive capabilities of new models of artificial intelligence, such as Claude Mythos or GPT 5.5, this can become our daily life tomorrow […] Within a year, there are dozens of models open source which will make these capabilities accessible everywhere, at low cost, without any safeguards.
If it is true what Anthropic claims – Mythos being able to detect all cyber security vulnerabilities – then it could pose a major risk if this AI system falls into the wrong hands. […] if we understand the experts correctly, Mythos is just the first model, but many will follow soon.
What is new, however, is that AI has become the crucial weapon of war. Every day, authoritarian regimes try to undermine our democracies through cyber-attacks, disinformation and attacks on critical infrastructure. What used to take a bunch of talented hackers a month now only takes a few minutes.
We are facing an unstoppable and dizzying technological revolution […] The precautionary principle practised by the institutions and the European Parliament cannot lead us to fear, but collaboration with these cutting-edge start-ups cannot lead us to accentuate dependency either. […] against the ability of artificial intelligence to exploit the vulnerabilities of open societies.
Madam President, artificial intelligence is profoundly transforming the way we live, work and communicate, but it is also changing the landscape of threats to the security of the European Union. Today, the risks are no longer limited to traditional cyberattacks. AI systems can be used to automate large-scale attacks, manipulate information, create deepfakes highly credible and exploit critical vulnerabilities with unprecedented speed.
Two Big Techs The United States has launched sophisticated artificial intelligence systems that have proven capable of detecting and attacking points of vulnerability without any human intervention. The cybersecurity of hospital systems, power grids and other infrastructure is at risk. […] This is not, therefore, the time to deregulate or simplify […] We need effective regulators with more means to inspect and license.
A new arms race has begun – not with tanks, not with missiles, with algorithms. If real, Mythos is an algorithm of mass destruction, and it is unfolding now. The line between civilian innovation and instrument of war is disappearing before our eyes. […] EU must initiate a global AI governance regime to address the topic.
Several new AI systems are especially good at detecting and exploiting system vulnerabilities like they were superhackers on steroids. […] those who treat AI development as an arms race are foolish, dangerous and may lead to drastic consequences for humanity. The only sensible thing to do, as in the previous dangerous arms race, is to agree internationally on how to oversee the development of AI.
In April, in fact, Anthropic announced Mythos, an AI model that can autonomously find thousands of critical vulnerabilities in every computer system, so dangerous that it cannot be put on the market. The French company Mistral claims to have equivalent capabilities today.
The danger of what we have learned from Mythos research and the Glasswing arrangement is that AI-assisted cyber operations may compress months of expert vulnerability research into days. […] I also ask the presidency to activate the integrated political crisis response for AI-driven cyber threats. This should get the Union to be ready when the next mass incident hits our hospitals or schools. That is what the framework was built for. Let us use it.
We stand at the brink of a potential cyber bloodbath. AI models can now hack any system on a large scale and with the speed of light. Criminals, nation states, will use these AI models against our banks, our supermarkets, and all of our society will be tested. […] there is very limited time to build resilience – months at best.
I believe that the biggest challenge facing the European Union today is how to adapt our societies to artificial intelligence, how to have artificial intelligence at the service of our democracies […] Artificial intelligence and threats are profoundly transforming our societies, security, the economy, the functioning of our democracies.
Cybersecurity is a strategic economic and security priority, as AI makes attacks faster, larger and more autonomous. Advanced AI models can identify and exploit hidden software vulnerabilities and therefore mark a turning point for European cybersecurity. At stake are public safety, economic stability and democratic resilience. Europe must act pragmatically. We need a dedicated EU strategy for AI-driven cyberthreats.
And advanced AI systems are transforming many sectors of our economy, from healthcare to transportation, finance, energy, education and defence. When it comes to cybersecurity, the new generation of advanced AI systems are powerful assets and, at the same time, pose significant risks. […] with the Artificial Intelligence Act, we already have a solid legislative framework that will allow us to monitor the most powerful AI models that pose systemic risks.
These realities can bring enormous benefits to humanity, but they carry equally disproportionate risks: Disinformation, algorithmic amplification, the manipulation of opinion and an all-powerful AI without ethical barriers endanger democracy, the critical thinking of human beings and even the freedom of the person.
Are we going to AI heaven where we all live like Greek gods served by AI robots, or AI hell where we are ruled by techno‑feudal overlords? Probably neither, but we are witnessing the start of a civilisational shift. The rise of artificial intelligence will shape our economies, our societies, and our future.
Anthropic announces Claude Mythos
Anthropic announces Claude Mythos, its most capable model to date, with leading performance on cybersecurity and biology tasks and access initially limited to vetted partners.
Can it be accepted today that artificial intelligence services are producing massive amounts of pedophile content of unprecedented violence? In 2025, 1,286 videos were generated by AI, including on X, thanks to Grok. […] Your idols - Musk, his MAGA clique - built with Grok a weapon, a real weapon.
2025
Many experts anticipate that the development of so-called ‘Artificial general intelligence’ (AGI) – systems with cognitive skills at or above human level – could become a reality within the next five to ten years, i.e. by 2035. A landmark technological development of this kind is likely to have far-reaching effects on the European labour market, public budgets and the EU’s economic competitiveness.
The exponential growth of AI is unlike anything we've ever seen in human history. […] Falling behind in the AI race would be like missing out on the Industrial Revolution 100 times over.
Artificial intelligence models that are capable of creating synthetic child sexual abuse material. AI-generated nude pictures of children are being produced from users accessing readily available images from online platforms. […] this technology is becoming increasingly accessible and tailored for abusers to bypass detection and regulation.
OpenAI’s new artificial intelligence model ‘o3’ refused to shut down when ordered to do so by its creators […] the model altered its own code to circumvent the shut down process […] these systems are trained with huge amounts of data and have the potential to autonomously create ways to achieve objectives without human control, questions arise as to their compatibility with the principles of security, accountability and fundamental rights.
AI will replace most jobs and we are not ready for it. Self-driving cars are replacing drivers. AI models are replacing software developers. Robots are replacing factory workers. Jobs will be disappearing, and the hard part is that we cannot really predict which ones will disappear and when.
2024
OpenAI announces o3
OpenAI announces o3, a reasoning model trained to work through a problem step by step before answering, aimed at harder maths, science, and coding tasks.
Innovation is always an opportunity, but only if you are able to steer it and you have to do it before it makes you obsolete. […] This AI act is our timely attempt to control the future developments before they control us.
We haven’t yet witnessed the full power of AI. The existing capabilities have already surpassed anything we could have imagined a decade ago, at speeds we have never thought possible, and this will continue exponentially. For lack of a better term, artificial general intelligence, or AGI, is something we need to prepare for. Any AI possessing an intelligence that is superior to that of a human being will open up infinite possibilities, but at the same time, will raise never before encountered ethical, moral, and yes, even existential questions. We have in the act a first answer to those questions, but we must be prepared to cater our governance for quantum leaps that we know AI could do.
2023
We now witness the meteoric development of generative AI. We know it will have a massive impact on the world as we know it, and we also know it is only the beginning of this technology. It will keep on leaping many times over. […] It is no doubt the most disruptive and consequential technology in our history to date.
We all use big words. We said change the world. Transformative impact. Turning points for humankind. Oftentimes we use these words as political exaggerations, but with AI, these are right on point. We are facing a turning point in history. Many may not see it yet. And we needed a small demonstration provided by a powerful chatbot over the past months to alert leaders and policymakers around the world.
As the godfather of AI, Geoffrey Hinton has said, AI also presents an existential threat to humanity, or in the vivid words of Yuval Noah Harari, it has the potential to hack the operating system of our civilization.
Half of all AI researchers said in a 2022 survey that there is a 10% chance that AI could lead to the extinction of humanity. Which of you would still get on a plane in such a probability of a crash? […] This is precisely why there are rules on aviation safety so that we can use this technology safely.
Artificial intelligence is like a river. We can put a dam in it to block it, but sooner or later it will overflow. What we have to do is try to direct the course. […] artificial intelligence is not neutral, it is a human product and in the management of these systems must never lack the control of the human being. […] the norm is not enough for the revolution that is coming.
It is right and proper to want to create a regulatory framework that guarantees the safe use of this instrument, that can offer great opportunities but that brings with it a huge destructive potential. We have heard many distinguished personalities from the world of technology raise doubts about the safety of this tool. […] without forgetting that the implementation of these technologies could lead to the disappearance of millions of jobs.
The digital revolution – and even the cognitive revolution – is on its way […] Experts predict that the machine will surpass human intelligence in 2029 and that by 2045 it will be a billion times smarter than all human brains.
Hoping that companies will self-regulate is not enough to safeguard our citizens in our societies. And we are now hearing more and more of the digital giants out there fearing their own products. […] What took me by surprise and by storm is generative AI based on powerful foundation models. […] ensuring that the models potentially used by millions of smaller companies and start-ups, as well as by hundreds of millions of individual users, that they are safe.
The whole world is watching us as alarms follow one another about artificial intelligence in general, the risk of extinction of the human race by sentient algorithms, alarms often launched by the same Big Tech that fear what they have created.
2022
ChatGPT is released
OpenAI releases ChatGPT, a chatbot built on the GPT-3.5 model that holds back-and-forth conversations and answers follow-up questions.