
Tom's Hardware
US hardware and technology journalism site; challenged Anthropic's Mythos zero-day claims, finding they rested on only 198 manual reviews.
Last refreshed: 16 April 2026 · Appears in 1 active topic
Did Tom's Hardware's Mythos critique hold up when AISI published its independent evaluation?
Timeline for Tom's Hardware
Mentioned in: AISI confirms Mythos 20-hour attack chain
AI: Jobs, Power & Money- Did Tom's Hardware debunk Anthropic's Mythos zero-day claims?
- Tom's Hardware found the 'thousands of zero-days' claim rested on only 198 manual reviews, many in outdated software. AISI's subsequent evaluation confirmed no single-task superiority for Mythos but validated a genuine 20-hour autonomous attack-chaining capability.Source: Tom's Hardware / UK AI Security Institute
Background
Tom's Hardware is a US-based technology journalism publication specialising in hardware reviews, benchmarks, and technical analysis. It covers processors, graphics cards, memory, storage, and related technology with a readership focused on enthusiasts and professional IT. In April 2026, Tom's Hardware published a critical review of Anthropic's marketing claims around Claude Mythos Preview's zero-day vulnerability discovery, noting that the 'thousands of zero-days' assertion rested on only 198 manual reviews and that many flagged vulnerabilities were in outdated software no longer in active use.
The Tom's Hardware critique arrived ahead of the UK AI Security Institute's independent evaluation on 15 April 2026, which reached a more nuanced conclusion: Mythos shows no single-task superiority over public frontier models on CTF benchmarks, but has a genuine advantage in multi-step autonomous attack chains. The AISI finding partly vindicated the Tom's Hardware scepticism about the zero-day count — the marketing claim was contested — while confirming the underlying attack-chaining capability was real.
Tom's Hardware's role on this beat is as an independent technical check on AI safety marketing claims. The publication is not a policy actor but its April 2026 review is cited in Anthropic's own Alignment Risk Update footnotes as part of the contested evidence record. That a hardware-focused trade publication was the first to scrutinise the Mythos claims is itself a signal: the specialist AI safety community was not publicly contesting the numbers before AISI published.