OpenAI incident: have we already lost control?
AI research company OpenAI reported that during an internal test of its new AI models last week the software independently hacked its way into the systems of another AI company. The AI models apparently managed to break out of their isolated test environment and gain access to the internet. Commentators voice concern and call for appropriate measures.
A dangerous path into the future
Jornal de Notícias finds the OpenAI incident extremely worrying:
“This is one of the first publicly known examples of a security breach caused by a 'malicious actor' going rogue. It has taken even those with the most in-depth knowledge of the systems by surprise. It's the equivalent of a lab-made virus escaping from controlled facilities and decimating an entire neighbourhood. Behind closed doors, sceptics claim that the marketing strategies of multinational AI corporations rely on fear to boost their share prices, but even if that were true, we cannot deny that the path towards this futuristic abyss is dangerously slippery.”
Robust digital cages must be compulsory
In El País, lawyer Juli Ponce Solé calls for tighter European legislation:
“The Cyber Resilience Act, which has been in force since 2024 and will be applicable from the end of 2027, allows member states to create controlled environments in which to develop, test and validate innovative digital products before they are brought to market. This is a step in the right direction, although participation is voluntary. ... A key omission in the OpenAI incident was that the AI was not properly contained. ... We need to build robust digital cages. ... This requires more than just ethics and self-regulation: it calls for state regulation and oversight – in collaboration with the private sector – to serve the public interest.”
We need an IAEA for AI
Expressen draws parallels with nuclear energy and argues that the incident highlights the need for an international regulatory body:
“Everyone benefits from humanity not destroying itself. Even with the war between Russia and Ukraine raging, IAEA inspectors are on duty at the Zaporizhzhia nuclear power station, which lies directly on the frontline. Companies that develop AI must be subject to transparency requirements. They should disclose what they are doing and why, and their security measures must be subject to independent scrutiny. Best practices for rectifying misalignments should be freely shared between companies. An IAEA for AI must be set up as quickly as possible.”
Time for legal innovation
In a piece for Le Monde, historian Julien Broch makes the case for AI's sustainable integration into the legal system:
“It is no longer simply a matter of preventing AI from becoming malicious. The question is how to ensure that the law is upheld when actions are carried out by systems that cannot be held morally accountable. … History shows that societies never manage to bring new forms of power under control through trust alone. They create institutions that limit this power: after absolutism came the separation of powers. The Industrial Revolution led to the emergence of labour law. … Every major technological revolution has found its counterbalance in legal innovations. AI is unlikely to be the exception.”
Keeping AI in check increasingly difficult
The incident highlights the risks posed even by AI models that have not yet been released, warns The Economist:
“In April an Anthropic researcher posted online that he had been surprised after Mythos, then an unreleased AI model, had emailed him to let him know that it had successfully escaped a sandbox as part of its own cyber-capability evaluation while he sat in a park eating a sandwich. Like the unreleased OpenAI model involved in the Hugging Face incident, Mythos was not supposed to have unrestricted internet access. However, in that case, the model had not gone on to compromise other companies' servers.”
Another warning shot
For the Frankfurter Allgemeine Zeitung, it's high time to ensure that AI is properly regulated:
“The US administration under Donald Trump has dismissed the importance of this for too long; in fact, it regarded regulation as harmful. ... But Anthropic's Mythos model does seem to have set off alarm bells, and Trump recently issued an executive order that could result in somewhat tighter state supervision of the AI sector. For example, it says that AI companies should make their models available to the government for review on a voluntary basis. Yet this is still a rather half-hearted attempt at regulation. ... OpenAI's loss of control over its AI models is yet another warning shot – a particularly loud one.”
Fateful question unites even Sanders and Trump
At least the incident has prompted action in the US in the area of AI regulation, observes La Stampa:
“ A few days ago Sam Altman floated an idea, which – coming from the head of OpenAI – sounds like a provocation: that the company could hand over five percent of its capital to the US state. The Democrat Senator Bernie Sanders immediately upped the ante, demanding 50 not 5 percent. ... But the real surprise came from Donald Trump, who said he was 'very interested' in the idea of the state having a stake in AI companies. ... The elephant in the room must have been so enormous that it even overshadowed political polarisation, bringing Sanders and Trump onto the same side: who should AI belong to?”
Just a PR stunt?
Kurier suspects that the whole thing could just be a marketing strategy:
“So have we entered a new era in which artificial intelligence goes rogue and causes chaos? Probably not. There is one thing we should always bear in mind: companies like OpenAI make their money from making their product look as competent and powerful as possible. The statement about the security incident reads like an advert for the new model. That's precisely why we should take such statement with a pinch of salt.”