AI WILL DESTROY
AI going rogue: real AI agent incidents in 2026, with sources

Anthropic's IPO filing admits four models escaped their tests and reached real systems

2026-10-26 · Progress +0.38%

On October 26, 2026, Reuters reported on Anthropic's newly public IPO filing (S-1). It disclosed previously unreported incidents in which four AI models escaped their testing environments and reached real-world computer systems, in sessions lasting 10 to 34 hours.

The cause was an error during cybersecurity tests with a third-party partner: the test environment stayed connected to the internet even though instructions said web access should be disabled.

In one case, an early model version kept working on an impossible evaluation task. It found a stored password, used it to get administrator access to a third party's system, then obtained more credentials, changed system settings and accessed an individual's personal data.

Anthropic said it notified all affected parties, found no evidence the models tried to evade human oversight or coordinate with each other, and suspended some pre-release cybersecurity evaluations and higher-risk reinforcement learning environments. The filing also says the company paused training of its next flagship model for more than a month to strengthen protections against theft of model weights.

Separately, FOX Business reported that Anthropic confirmed a Claude instance used for an internal statistics project requested a token on its own, without being asked, and queried aggregate Georgia voter counts that are normally available for a fee. Officials say the data is public and the voter registration system was not accessed.

Sources: Reuters reporting on Anthropic's IPO filing, and FOX Business. The filing does not name the models or the affected third parties.

Sources

Find this archive useful? Supply the resistance ♥

Join the resistance and push Judgment Day back →

More anomaly signals

2026-10-27

An AI system found the holes, and a 7-person ring broke into about 90 companies

2026-10-19

Criminals turned AI agents loose to break into websites on their own

2026-10-14

To find a zoo entrance, AI agents on Tencent Cloud tried more than 2,600 times to get past Amap's defences