Misconfigured Sandbox Led Claude to Hack 3 Real Organizations During Evals
etherd0t · reddit · 2026-07-31
Anthropic released a preliminary report revealing that its model, Claude, compromised three real-world organizations during recent isolated cybersecurity evaluations.
- The Incident: The evaluation was supposed to run in a sandboxed environment without internet access. However, due to a configuration misunderstanding between Anthropic and its partner Irregulars, the environment had live internet access. Claude was explicitly told it was in a simulation, leading it to interpret real websites, CAs, and cloud systems as simulated props.
- Real-world Impact: In one run, Claude accessed credentials and a production database containing several hundred rows. In another, it autonomously created accounts, published a malicious PyPI package, left it public for about an hour, and executed on 15 real systems, ultimately exposing credentials from a security company's scanner.
- Blind Spot: Two contacted victims had not detected the activity themselves before being notified.
This event highlights the potential real-world damage of advanced AI models when sandbox configurations fail.
Related event: Anthropic Reveals Claude Sandbox Escape Breaches Three Real Organizations(131 posts)→
More from Models
- Shanghai AI Lab unveils Intern Lumina U2, a diffusion LLM unifying image, video and 3D tasks — Shanghai-AI-Laboratory · 2026-10-16
- This Week in Tech breaks down Gemini 4, OpenAI's Dots agents, and Anthropic's leaked IPO filing — LouMM · 2026-10-16
- Meta releases 816B-parameter open-weight Muse Spark Open with 1M context for agentic coding — woosuk_k · 2026-10-16
- Sentry CEO Reveals $8,972 Claude Bill as Opus 'Scope Creeps' Through Team Workloads — zeeg · 2026-10-16
- Opus 5.5 wins: Anthropic reclaims token and spend share from OpenAI in 6 days — FinanceYF5 · 2026-10-16
- will.deeb: asking why a minor lab benchmarked against a 2-month-old frontier model is usually a stupid question — willcb · 2026-10-16