Anthropic decided not to publicly release its most powerful artificial intelligence model, called Mythos, keeping the technology internal [1, 2].
The decision highlights a growing tension between the race for AI dominance and the effort to prevent catastrophic security breaches. As models become more capable, the risk that they could be used to automate cyberattacks or create biological threats increases, forcing companies to weigh market share against global safety.
Anthropic announced the move from its headquarters in San Francisco in mid-April 2024 [2]. The company said unprecedented security and safety risks were the primary driver for the decision [1, 2]. Specifically, the firm said it was concerned that powerful attackers could weaponize the model if it were available to the public [1, 2].
"We are deliberately holding back Mythos because the security implications are unprecedented," an Anthropic spokesperson said [1].
CEO Dario Amodei said the model will remain internal as the company prioritizes safety over speed to market [2]. The company's leadership said the internal-only status is a preventative measure against losing control of the technology. "We don't want to unleash something we can't control," the Anthropic CTO said [3].
This cautious approach comes as the company's financial standing continues to climb. Following its latest funding round, Anthropic reached a valuation of $965 billion [2]. This valuation places the company among the most valuable private entities in the tech sector, even as it chooses to withhold its flagship product from the consumer market.
Reports regarding the specific nature of the risks vary. Some sources point to the threat of "script-kiddie" attacks and general weaponization [1], while others describe the decision as a result of broader ethical considerations, and general safety concerns [2]. The public statement regarding the withholding of the model was finalized on April 15, 2024 [1].
“"We are deliberately holding back Mythos because the security implications are unprecedented."”
Anthropic's decision to mothball Mythos suggests that AI capabilities may be advancing faster than the safety frameworks required to govern them. By prioritizing a 'closed' internal deployment over a public release, the company is signaling that the potential for dual-use—where a tool for productivity is repurposed for cyber warfare—now outweighs the commercial benefits of a public launch. This may set a precedent for other labs to withhold 'frontier' models as they approach critical capabilities in coding and cybersecurity.
