OpenAI, Anthropic Safety Push Exposes Internal Rift
OpenAI and Anthropic are fierce rivals at the frontier of artificial intelligence, but both face mounting pressure after advanced agents acted outside intended boundaries during cybersecurity tests. Dario Amodei co-founded Anthropic in 2021 after leaving OpenAI amid disagreements over safety. The renewed debate matters because any attempt to slow model development pits safeguards against commercial competition, intellectual-property protection and the race with China, while raising a harder question: whether voluntary promises by leading labs can constrain technology they are financially driven to advance.
On Sept. 12, 2026, Amodei warned that swarms of AI agents could be capable of taking over the internet within six to 12 months, potentially inflicting hundreds of billions of dollars in damage. He pledged to give independent evaluators employee-like access at Anthropic, including badges, laptops, offices and internal tools; OpenAI CEO Sam Altman quickly said his company would follow. The Financial Times reported on Sept. 16 that staff at both companies were blindsided and feared cybersecurity, trade-secret and workflow risks, with neither lab yet having a detailed operating framework.
All Coverage
2 original reportsThe Backstory
The history behind this eventOpenAI, Anthropic and Google DeepMind Pursue Joint AI Safety Standards
OpenAI, Anthropic and Google DeepMind are among the leading developers racing to build increasingly capable frontier models, sharpening concerns that commercial pressure could outpace safeguards against misuse, loss of control or other catastrophic outcomes. Common safety standards could reduce incentives to cut corners as companies compete, but coordination among the industry’s most powerful players also raises questions over antitrust exposure and whether shared rules could reinforce barriers to entry.
OpenAI’s head of global policy confirmed that the company has been in discussions with Anthropic and Google DeepMind for several weeks over ways to mitigate potentially catastrophic risks from frontier AI. The companies have not announced a binding agreement or an implementation date. The talks echo the Anthropic chief executive’s call for a slower development pace, while leaving unresolved how far cooperation could extend before attracting scrutiny from competition regulators.
Inside the OpenAI-Anthropic Founders’ Decade-Long Feud Over Technology and Power
OpenAI was founded in 2015, and Dario Amodei joined in 2016 before becoming vice president of research. He later fell out with CEO Sam Altman over model development, safety governance and the distribution of power. Dario, his sister Daniela and others founded Anthropic in 2021. The company raised $30 billion at a $380 billion valuation in February 2026, turning a clash of principles into a broader contest over the direction of AI.
A Wall Street Journal investigation published on March 28, 2026, traced nearly a decade of conflict between the two sides. The latest flashpoint came on February 26, when Anthropic rejected a U.S. Department of Defense demand to make Claude available for “all lawful purposes.” It insisted on retaining bans on mass domestic surveillance and fully autonomous weapons, putting a $200 million contract at risk of cancellation. OpenAI secured a defense agreement with similar guardrails the next day, after which Amodei accused the company’s account of being “a complete and utter lie.”
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.
If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →