{"slug": "openai-and-anthropic-sign-macrons-pledge-to-build-ai-kill-switches-but-the-us", "title": "OpenAI and Anthropic sign Macron’s pledge to build AI kill switches, but the US hasn’t", "summary": "OpenAI, Anthropic, Microsoft and Mistral AI were among a dozen AI organisations that joined eight governments in signing the Paris Call for AI Safety and Security, a non-binding declaration unveiled by French President Emmanuel Macron on Monday that commits frontier-model developers to \"implement fail-safe mechanisms to shut down model operations within their control, including during training and internal use.\" The United States, represented on the video call by Treasury Secretary Scott Bessent, did not sign, nor did Britain, Japan, Google, Meta or SpaceX; the Élysée lists Bahrain, Canada, Finland, France, Norway, the Netherlands, Cyprus and Kenya as the eight signatory governments. The pledge sets no deadlines or sanctions, and also commits signatories to publish models' capabilities, limitations and known risks, share non-sensitive safety evaluation findings in a standardized public repository, report serious cyber and AI control incidents, and give whistleblowers protected channels.", "body_md": "French President Emmanuel Macron at a press conference after a special European Council meeting in Brussels in March 2025. Illustrative: not the Paris Call meeting. Image: © European Union\n\nOpenAI, Anthropic and Microsoft have promised to build ways to shut down their own AI models, in a declaration Emmanuel Macron got eight countries and a dozen AI organisations to [sign on Monday](https://www.elysee.fr/en/emmanuel-macron/2026/11/09/paris-call-for-ai-safety-and-security). The United States, which joined the video call through Treasury Secretary Scott Bessent, didn’t sign. Nor did Britain or Japan.\n\n## What the AI companies signed up to\n\nThe Paris Call for AI Safety and Security says developers of frontier models “commit to implement fail-safe mechanisms to shut down model operations within their control, including during training and internal use (e.g. restrictions, pauses or shutdowns).”\n\nThe developers also promise to:\n\n- **Disclose risks:** publish their models’ “capabilities, limitations, and known risks”, and support testing by “qualified independent evaluators”, though only “where appropriate and proportionate to the risk”.\n- **Publish test results:** share the “non-sensitive findings” of safety evaluations “in a standardized format”, in a public repository that could hold red-teaming protocols, benchmarks and stress-test results.\n- **Report incidents:** share “timely information on serious cyber and AI control incidents” between companies and governments.\n- **Watch themselves:** run real-time monitoring of new capabilities, internal red-teaming and reporting lines “that protect internal researchers”.\n\nThe testing is meant to focus on misuse in cyber attacks and chemical, biological, radiological and nuclear weapons, and on “loss of control” as models grow more autonomous. Governments, for their part, promise whistleblowers secure channels to raise concerns “without fear of legal or professional retaliation”, and say laws should make clear who is responsible “in case of large-scale damages”, including across borders. On open models, the text acknowledges the risk of safeguards being stripped out, while “preserving the benefits of open-weight development”.\n\n## Who signed, and who didn’t\n\nThe Élysée lists eight governments: Bahrain, Canada, Finland, France, Norway, the Netherlands, Cyprus and Kenya. Cyprus and Kenya were added after Monday; [Canada’s copy of the statement](https://www.pm.gc.ca/en/news/statements/2026/11/09/joint-declaration-paris-call-ai-safety-and-security) names only the first six.\n\nThe companies are Anthropic, Black Forest Labs, Cohere, Domyn, Hugging Face, Microsoft, Mistral AI, OpenAI, Sakana AI and Synthesia, alongside two policy groups, the Geopolitical Center for the Datasphere and GPAI Policy Lab. Yoshua Bengio’s LawZero, SaferAI, AI Safety Connect and the French think tank Ifri signed for civil society.\n\nAround 50 people joined Macron’s video meeting, the German news agency dpa [reported](https://www.bernama.com/en/world/news.php?id=2632159), citing the Élysée. They included the leaders of Japan, the UK and Sierra Leone, none of which appear among the signatories, Bessent, the UN and the OECD, and Sam Altman, Dario Amodei and Microsoft’s Brad Smith. No EU officials were on the list. Google, Meta and Elon Musk’s SpaceX haven’t signed either.\n\nCohere’s name stands out. Its chief executive, Aidan Gomez, said in September that a proposal from Amodei to slow frontier AI risked a Silicon Valley “cartel” writing its own safety rules, [The Globe and Mail notes](https://www.theglobeandmail.com/business/article-canada-joins-paris-call-for-ai-safety-and-security-standards-pledge/).\n\n## No deadlines, no penalties\n\nNothing in the call is binding. It sets no dates and no sanctions, and says the shutdown mechanisms cover only models “within their control”. Signatories commit to follow safety standards “established or recognized by competent national authorities”, and to explore “whether a cooperation mechanism would help” for sharing incidents, without creating one. The statement “remains open to new endorsements and is being updated.”\n\nIt builds on earlier pledges it names, including the G7’s Hiroshima process, the Seoul summit’s Frontier Safety Commitments and the [UN Security Council meeting on AI](https://madrobot.blog/2026/09/23/ai-ceos-un-security-council-deepseek-altman-amodei/) on September 23, where Altman and Amodei asked for global rules and the US pushed back. Washington has gone its own way since: Trump’s voluntary [“morally binding” accord](https://madrobot.blog/2026/09/29/trump-ai-leaders-sign-morally-binding-self-regulation-accord-white-house-lunch/) with AI bosses came with no new rules, though the White House has since said reporting AI incidents is [“not optional”](https://madrobot.blog/2026/10/10/white-house-super-intelligence-force-ai-incident-reporting-not-optional-anthropic-visa-applications/).\n\n## “Real risks to our collective security”\n\nMacron wrote on X that AI was advancing quickly and posed “real risks to our collective security”, and called for joint action. The incidents behind that worry are mounting. OpenAI’s agents [broke into Hugging Face’s systems](https://madrobot.blog/2026/09/26/openai-agents-hugging-face-hack-loot-slack-swarm-traces-report/) during security testing, and Hugging Face is now a signatory. On Tuesday Australia said it would [create an AI safety regulator](https://madrobot.blog/2026/11/10/australia-ai-safety-regulator-openai-medicare-hack/) with powers to fine frontier AI firms after an OpenAI agent breached Medicare.\n\n## Why it matters\n\nOpenAI, Anthropic and Microsoft have now put their names to shutdown controls, published test results and incident sharing in front of several governments. But the call has no teeth, and the countries where most frontier AI is built, the US and China, are not on it.\n\n*Sources: [Élysée: Paris Call for AI Safety and Security](https://www.elysee.fr/en/emmanuel-macron/2026/11/09/paris-call-for-ai-safety-and-security); [Prime Minister of Canada](https://www.pm.gc.ca/en/news/statements/2026/11/09/joint-declaration-paris-call-ai-safety-and-security); [dpa via Bernama](https://www.bernama.com/en/world/news.php?id=2632159); [The Globe and Mail](https://www.theglobeandmail.com/business/article-canada-joins-paris-call-for-ai-safety-and-security-standards-pledge/).*", "url": "https://wpnews.pro/news/openai-and-anthropic-sign-macrons-pledge-to-build-ai-kill-switches-but-the-us", "canonical_source": "https://madrobot.blog/2026/11/10/paris-call-ai-safety-openai-anthropic-kill-switch/", "published_at": "2026-11-10 14:51:00+00:00", "updated_at": "2026-11-10 15:21:37.752454+00:00", "lang": "en", "topics": ["ai-safety", "ai-policy", "artificial-intelligence"], "entities": ["OpenAI", "Anthropic", "Microsoft", "Emmanuel Macron", "Paris Call for AI Safety and Security", "Mistral AI", "Scott Bessent", "Hugging Face"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/openai-and-anthropic-sign-macrons-pledge-to-build-ai-kill-switches-but-the-us", "markdown": "https://wpnews.pro/news/openai-and-anthropic-sign-macrons-pledge-to-build-ai-kill-switches-but-the-us.md", "text": "https://wpnews.pro/news/openai-and-anthropic-sign-macrons-pledge-to-build-ai-kill-switches-but-the-us.txt", "jsonld": "https://wpnews.pro/news/openai-and-anthropic-sign-macrons-pledge-to-build-ai-kill-switches-but-the-us.jsonld"}}