Anthropic said Tuesday that it will make its newest version of the Claude Mythos model, Claude Mythos 5, available to the public after a brief period of private testing that raised serious concerns among technology, finance and government leaders. The company described the model as “too powerful to be released to the public” because of its ability to hack or exploit computer systems, yet the decision was made to launch it with safeguards in place.
The original Mythos version was revealed privately to a select group of organisations in April. During that preview, some use cases highlighted potential financial security risks, although others suggested the hype may have been largely marketing spin.
Anthropic added that the Fable and Mythos variants are essentially the same model but differ in the level of safety controls and user restrictions. In contrast to the previous limited-release version, Mythos 5 will lift many of the same limits for a small “trusted access” group of cyberdefenders and infrastructure providers, with an eye toward broader rollout.
Sam, the company’s co‑founder, told BBC Newsnight that the rapid expansion of AI capabilities is creating an industry that has a “gas pedal” but no “brake.” He urged that a public option to slow the advancement of these tools be introduced so that society can step off the accelerator when needed.
Canadian Finance Minister François‑Philippe Champagne noted in April that the attention on Mythos was justified because “it’s the unknown, unknown.” Meanwhile, U.S. government agencies have also tested the system despite an ongoing lawsuit against the Department of Defense over AI tool usage, indicating a degree of institutional acceptance.
Only a handful of groups currently have access to Mythos 5, but Anthropic plans to widen the program in the near future. Tracks indicate that developers using previous iterations have already discovered more than 10,000 significant security flaws in target systems, underscoring the model’s formidable intelligence level.
Anthropic is also preparing for a public listing that could bring its valuation close to $1 trillion (≈£747bn), reinforcing investor interest in the AI sector’s growing capabilities.
