Anthropic Urges Slower AI Progress Amid Risks

Anthropic has signaled public caution about the pace of AI development while simultaneously restricting access to a high-risk internal model. According to ABC, Anthropic's blog post suggested that "if it were possible to slow the development of this technology ... we think that would likely be a good thing," conditional on a mechanism for global coordination. BBC reports Anthropic co-founder Jack Clark told BBC Newsnight, "You want the option to be able to take your foot off the gas and put your foot on the brake." ABC reports the company developed a model called Mythos, which Anthropic described as "strikingly capable at computer security tasks," and that access was limited to a select group of insiders and partners.

ABC reports that access to Mythos was confined to a small group, including about 150 partner organisations and a subset of Australian firms, so they could test defenses rather than release the model publicly. BBC reports Claude is operating on code the system wrote itself for roughly 80% of its codebase, and that reaching full self-authored code could be possible within two years, according to Jack Clark's remarks to BBC.

Editorial analysis: Public calls for a slowdown coexist with continued private development and selective sharing, a pattern seen previously when firms manage dual-use advances through controlled releases and security collaborations. Observed patterns in comparable situations show regulators, corporate partners, and security teams often scramble to translate advisories into operational tests and policy proposals before multilateral agreements can form.

Editorial analysis: The combination of high capability claims about Mythos and the reported autonomy of Claude on software generation raises two practitioner concerns: elevated operational risk from dual-use capabilities, and a greater need for robust red teaming and deployment gating. Industry attention will also focus on whether private-sector coordination models, like the reported partner program, scale effectively in the absence of global regulatory mechanisms.

Editorial analysis: Observers should track whether governments respond with concrete testing or reporting requirements, whether other labs adopt similar restricted-access programs, and whether independent red teams publicly reproduce Mythos-class capabilities. Also monitor investor and market signals, since BBC reports private valuations near $1 trillion, which can affect incentive structures across the sector.