Ten AI companies, including OpenAI, Anthropic, Microsoft and Mistral, have signed a new declaration on the safety of advanced AI models. The Paris Call for AI Safety and Security was agreed Monday at a meeting hosted by French President Emmanuel Macron. Critics say it does not go far enough.
Macron said artificial intelligence is advancing quickly and poses “real risks to our collective security.” He called on governments and companies to work together to address those risks. In a post on social media, he said the meeting had brought together around 50 people, including heads of state and government, leaders of major AI labs, researchers and experts, some in person in Paris and others by video.
The signatories include Hugging Face, Cohere, Black Forest Labs, Domyn, Sakana AI and Synthesia, alongside the four better-known firms. Six countries signed as well: France, Canada, Finland, Norway, the Netherlands and Bahrain. Yoshua Bengio’s AI safety organization, LawZero, is among the research groups supporting the effort.
OpenAI chief executive Sam Altman, Anthropic chief executive Dario Amodei and Microsoft president Brad Smith joined the discussions by video. US Treasury Secretary Scott Bessent also took part, but the United States is not among the signatories. Google, Meta and China have not signed either.
What does the Paris Call for AI Safety require?
Under the declaration, the companies have committed to building fail-safe mechanisms that can shut down their AI models, including during training and internal use. They also pledged to give clearer information about what their systems can and cannot do, and to allow qualified independent experts to test them where appropriate.
The document calls for a shared approach to assessing and managing the risks of frontier AI. Those risks include malicious uses such as cyberattacks and weapons development, as well as the possibility that AI systems could act independently or move beyond human control. Test results that are not sensitive would be published in a common format. Governments and companies would also share early warnings about serious incidents, and whistleblowers would get secure channels to report concerns.
Why do critics say the Paris Call lacks teeth?
CeSIA, which signed the declaration, has criticized it anyway. The group says the text sets no common risk thresholds for deciding whether a model should be developed or released. It also does not guarantee that independent testers or governments will have access to the systems they need to evaluate.
CeSIA wants governments to have the authority to suspend a model when its risks become too high. The group argues that developers need clearer obligations and a timetable for meeting them.
“These decisions must rest with public authorities,” said Charbel-Raphaël Ségerie, CeSIA’s executive director.
The Paris Call follows an initiative Norway and Finland launched in September to bring frontier AI under greater oversight. France later joined that effort. The declaration also builds on earlier safety commitments AI companies made in Seoul and on discussions at the G7 summit that France hosted in Evian.
The declaration remains open to other signatories, but it sets no deadlines. The shared testing framework it proposes has not yet been developed.
On Tuesday, the forum turned to a different concern for Macron: how AI affects children. Speaking at a summit on children in the AI era, he called for AI tools to undergo checks before they are made available to young users.
“We will push for pre-certification of all artificial intelligence solutions,” he said.
Macron compared the idea to the safety checks required for children’s toys. Parents would not give a child an untested toy, he argued, and AI tools should meet the same standard. He also warned that the effects of AI on young people’s mental health are already becoming visible, as more children and teenagers turn to chatbots that can influence their emotions and how they see the world. A charter on safer AI for children was scheduled to be signed later Tuesday.
Macron also received a report from a government-appointed panel of experts. Titled “Protect without preventing,” it examines generative AI and human vulnerability.
Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.