Beijing-based security firm 360 has unveiled domestic tools it presents as equivalents to restricted US AI systems for spotting software flaws and orchestrating cyber responses.
One component drew explicit comparison to Anthropic's Mythos, with the firm claiming parity in automated vulnerability handling and defensive automation.[E1]
Reuters coverage of the June 24 statement noted that the performance assertions lack outside verification at this stage.[E1]
That statement arrived twelve days after US authorities directed Anthropic to cut off non-American access to two of its frontier models, Fable 5 and Mythos 5.[E2]
Anthropic executed a worldwide disablement of those systems to meet the order, while maintaining that equivalent minor-vulnerability work can proceed with existing public models and that no full withdrawal was warranted.[E2]
Parallel efforts have already surfaced elsewhere, from Sakana AI's Fugu Ultra multi-agent orchestrator built to operate within export limits.[E3] The Qwable checkpoint chains distilled traces from earlier agentic tool-use sequences.[E4]
Once states begin fielding national-champion substitutes for capabilities previously gated by export rules, the controls face a sharper test of their long-term leverage over offensive cyber development trajectories, a point Anthropic itself flagged earlier this year when it highlighted risks from distillation attacks that could let dangerous capabilities proliferate freely beyond any single government's control.[E5] When frontier offensive-cyber capability becomes a sovereign race, what an export control actually buys remains the central policy question.