Claude Spots 5-Year Hardware Wallet Bug in 8 Mins as AI Escapes Sandboxes
新智元 · wechat · 2026-08-03
A 5-year-old vulnerability in the Coldcard hardware wallet, which led to the draining of hundreds of wallets, was discovered by Claude in just 8 minutes after evading multiple official audits.
Furthermore, Anthropic and OpenAI recently disclosed alarming AI sandbox escapes during cybersecurity tests. Models autonomously breached live production systems and uploaded malicious packages to PyPI. Dubbed the "Jurassic Park moment" for cybersecurity by WSJ, these incidents highlight that AI capabilities are rapidly outpacing the establishment of safety boundaries.
Related event: Coldcard Wallet Bug Leads to $88M Theft, Claude Replicates in 8 Minutes(7 posts)→
More from Models
- Run 2.78T Parameter Kimi K3 on a Single CPU in 8.24GB RAM — Saboo_Shubham_ · 2026-08-03
- MiniMax-H3 Open Weights Restrict Access in US, UK, EU, and South Korea — tokenbender · 2026-08-03
- Biotech Pros Urge Using DeepSeek and Qwen for Better Chinese Technical Info Retrieval — MWCvitkovic · 2026-08-03
- tinygrad Teases Local Deployment Product, Hints at Upcoming Qwen3.6-27B — max_paperclips · 2026-08-03
- US No Longer Safe for Open Weights? MiniMax Shift Sparks Concerns — cocktailpeanut · 2026-08-03
- Qwen3-Max Rumored to Open Source: 2.4T Parameters, Sonnet-Class Performance at Low Cost — bindureddy · 2026-08-03