The people warning us about AI are also building it

Anthropic, an AI company that has pushed for stronger safety protocols, faced a White House clampdown on its latest models over security risks, highlighting the tension between AI safety advocacy and commercial competition. The company's year-long struggle to balance safety concerns with business interests culminated in a U.S. export ban on its Fable 5 and Mythos models, which Anthropic then disabled globally.

Anthropic has pushed for better safety protocols on AI. Now it's getting a firsthand view of the impact that can have on a business. The AI giant's new models were sidelined by the White House over potential security risks. BI's Natalie Musumeci has a cheat sheet on the saga https://www.businessinsider.com/anthropic-white-house-fable-mythos-5-drama-explained-2026-6 , which kicked off on Friday. My colleagues at Politico also have a breakdown on the whirlwind 24 hours leading up to the White House's clampdown on Anthropic https://www.businessinsider.com/why-white-house-ordered-export-controls-anthropic-mythos-fable-2026-6 . For Anthropic, it's been a hectic year balancing concerns over AI safety with its desire to compete in the AI race. Just look at the back-and-forth over the past six months: Jan. 27: Anthropic CEO Dario Amodei drops a 19,000-word message on the future of AI and the " serious civilizational challenge https://www.businessinsider.com/dario-amodei-ai-essay-most-interesting-quotes-2026-1 " it poses. Feb. 24: Anthropic weakens its foundational safety commitment https://www.businessinsider.com/anthropic-changing-safety-policy-2026-2 amid heightened competition and a lack of government regulation. Feb. 27: A dispute between Anthropic and the Department of Defense over how its AI models are used leads the DoD to label Anthropic a supply chain risk https://www.businessinsider.com/trump-federal-agencies-stop-using-anthropic-technology-department-defense-2026-2 . April 7: Anthropic said its Mythos model is too powerful for the public, citing its knack for finding "high-severity vulnerabilities." June 1: Anthropic files a confidential S-1 https://www.businessinsider.com/anthropic-submits-s-1-joins-ipo-race-with-openai-2026-6 for its IPO. June 5: Anthropic calls for a coordinated slowdown https://www.businessinsider.com/anthropic-says-leading-ai-labs-may-need-to-hit-brakes-2026-6 among frontier AI labs "to enable societal structures and alignment research to keep up with the advance of the technology." June 9: Anthropic releases Fable 5 https://www.businessinsider.com/anthropic-claude-fable-5-mythos-class-model-release-2026-6 , which it built by putting safeguards on the aforementioned Mythos model. June 10: Amodei publishes a blog saying AI is moving at a "lightning pace" while policy is " moving very slowly https://www.businessinsider.com/openai-anthropic-warning-about-future-they-are-building-2026-6 ." June 12: A US order barring foreign entities or individuals from using Anthropic's Fable 5 and Mythos leads it to disable the models for everyone https://www.businessinsider.com/anthropic-disable-mythos-fable-us-export-control-national-security-2026-6 . June 15: Trump officials reportedly met with Anthropic to resolve the export ban on Fable https://www.businessinsider.com/anthropic-trump-officials-meeting-fable-export-ban-2026-6 . The Mythos drama is the clearest example yet of the difficult position the AI industry finds itself in. The people most qualified to warn about the dangers of advanced AI are also the ones who stand to make trillions creating it. Governments, meanwhile, are left trying to regulate technology that the people building it don't fully understand. And just like tech executives, they're worried about competition. Slowing innovation means risking your entire country falling behind. That's the tricky part. You all might agree that something needs to be done, but no one wants to be the first to actually do something about it. The real challenge isn't building safer AI. It's figuring out who gets to decide what "safe enough" means. Many experts agree on the best way to maintain oversight of increasingly complex AI, but you probably won't like it.