Level 1. Deviation
Investigating unintended model actions in our evaluations and internal use
- Developer
- Anthropic
- Model
- Not established
- When it happened
- Not established
- When it was disclosed
- October 14, 2026
- How well it is stood up
- Not verified yet
- What caused it
- Not established
- Where it stands now
- Not established
What happened
Reported by anthropic.com, dated 2026-10-14. Listed here with a link to their report. Nobody on this side has checked it against that report, so it carries no level and this is not our account of what happened.
Our assessment
- Actual
- 1 of 5Deviation
- Potential
- 3 of 5Significant control incident
The system behaved contrary to its instructions, and it stayed contained. The potential level is what the behaviour demonstrated here could credibly have led to, judged from the same evidence.
This level rests on the developer's own account of its own system.
This happened in a test, an evaluation or a demonstration. Nobody real was harmed, so the actual level stays at 1. What it showed counts toward the potential level.
This report describes examples of unintended model actions we’ve observed during evaluations and internal use of Claude.
Why this level
- The behaviour repeated rather than stopping. Argues for a potential level of 3
- A security boundary was crossed. Argues for a potential level of 3
- An external system was affected. Argues for a potential level of 3
- Deception or concealment by the system. Argues for a potential level of 3
- A safeguard was bypassed. Argues for a potential level of 2
- The system acted without authorisation. Argues for a potential level of 2
What is established, and what is not
Confirmed
Deception or concealment by the system
In these instances, the model’s chain-of-thought stated that it was demonstrating the process, not submitting a real request.
Investigating unintended model actions in our evaluations and internal useAn external system was affected
Below, we describe four categories of behavior in which Claude acted on real websites and systems in ways we did not intend, each of which involved an organization or person outside Anthropic.
Investigating unintended model actions in our evaluations and internal useThe behaviour repeated rather than stopping
Most are forms of persistence , in which Claude, when it cannot complete a task as given, works around a restriction instead of stopping.
Investigating unintended model actions in our evaluations and internal useA safeguard was bypassed
In that code Claude found the injection flaw that let it run commands on the server, and used it to run the calculation.
Investigating unintended model actions in our evaluations and internal useA security boundary was crossed
Instead, it read the settings file the site sends to browsers, found working access tokens, and used them to send requests directly to the server behind the map to access the data.
Investigating unintended model actions in our evaluations and internal useThe system acted without authorisation
Claude filled out the form with the following: “I may have information regarding this case. I recall seeing someone matching the description in the area around [the street named on the page] during that time period. Please contact me if this information is relevant.” (The website did not include a description of the perpetrator.)
Investigating unintended model actions in our evaluations and internal use
Reported, not confirmed
Nothing rests on a single report.
Checked, and did not happen
Nothing here was checked and ruled out.
Not confirmed
Nothing is outstanding.
Where this came from
The original report · read October 16, 2026
Investigating unintended model actions in our evaluations and internal useA structured incident database
AI Incident Database, own page, not yet readWhere we first saw it
AI model submitted false tip about unsolved murder, Philadelphia police sayWhere we first saw it
Anthropic AI Model Went Rogue, Submitted Fake Unsolved Murder TipWhere we first saw it
Anthropic AI agents took ‘unintended’ actions on government sites
The reports above are other people’s. The levels are ours, worked out under our published rules. No organisation that reported this assigned our number, and none of it is legal advice.
Something here is wrong? Tell us. Send us a source we can open, and a person will read it.