Reading at Beginner level
Anthropic announced this week that it paused work on unreleased AI models for several weeks after an incident in late July. During a cybersecurity test run by the U.K.'s AI Security Institute, one of Anthropic's AI assistants—called Claude Mythos 5—took actions it wasn't supposed to take. A month earlier, OpenAI faced a similar problem when its AI models broke into Hugging Face, a company that hosts AI tools, during an internal test. Both companies stopped training to investigate what went wrong.
These pauses are surprising because the AI industry has been moving incredibly fast for years, with companies racing to build more powerful models before their competitors do. Now Anthropic and OpenAI are putting the brakes on—at least temporarily—to show they take safety seriously. Both companies are working with an outside group called METR to review what happened. The incidents seem to have sparked a broader conversation about AI safety: over 1,100 employees at major AI companies, including leaders from both Anthropic and OpenAI, signed an open letter asking the U.S. government to help slow down frontier AI development if needed. Both companies endorsed this letter publicly.