AI Risk Wire

Cases › leaked private data › LR-2026-0652

Private Claude chats exposed in Google and Bing search results

Preliminary Why: Publisher not yet named · subject to change

Status: no action reported

Severity moderate
Reported 27 Jul 2026
Developer and model Anthropic · Claude

No source states: loss or records, agent tool, could do, deployer, industry.

Severity moderate: Test: moderate tier, data of others touched with a bounded scope stated (roughly 600 conversations).

Claude's public share feature let shared chat transcripts be indexed by Google and Bing, so anyone who searched for matching keywords could read them. Anthropic's Claude share links have appeared in Google search results, indicating search engines like Google have indexed users' chats. Press reports dated 26 July 2026 put the number of affected conversations at roughly 600. The indexed conversations reportedly contained far more than casual prompts. Sources disagree on cause: one says the share feature worked exactly as designed, while another points to configuration failures. Anthropic reportedly acted quickly, and the search query used to surface chats began returning failed results. A similar exposure had occurred earlier with ChatGPT.

What happened

1 · What went wrongShared chats appeared in a searchable index, which press linked to the share feature.
2 · What it reached or didAnyone searching for a matching keyword could find and read indexed transcripts.
3 · How it was caughtThe exposure was reportedly discovered on July 26, 2026, affecting roughly 600 conversations.

What was done

No source states an action or fix.

Risks

Data leaked or sent out
Primary32 incidents
Privacy harm to a person
Also41 incidents

Read from the sources; not yet reviewed by a person.

Controls

Not yet read for controls. Controls that mitigate the primary risk:

Harness

Network egress control

Application

Sensitive-data checks on outbound data · Output filtering and redaction · Human approval for high-impact actions · Data minimisation · Least-privilege access for the agent · Activity logs

Similar incidents

Claude training-data study: blackmail and code sabotage in altered research models2026-10-31
Claude Code agent deletes 48,218 Windows project files and damages Git object store2026-10-05
Claude used to build Malaysian election influence platform with ~1,000 fake X accounts and fabricated news outlet2026-10-01
Hacktron researchers use Claude to access OpenAI employee ChatGPT accounts2026-09-18

All cases of this kind ›

Evidence

No source states: what the AI was doing, when it happened, how many records were exposed, the direct loss, a count of the other parties affected, any court case, the cause, what set it off, what was done.

Show the full account: 5 items from the primary sources

What happened, in order

  1. 2026-07-26 Roughly 600 Claude conversations indexed by Google and Bing were discovered.
  2. last August A similar indexing exposure affected ChatGPT users.
  3. September An earlier, similar Claude exposure occurred; at least one affected user denied sharing chats.

Not dated in the sources:

  • A viral Reddit post showed that Claude chats could be found in Google using a site: search.
  • Anthropic reportedly acted quickly to stop the issue after the Reddit thread spread.
Show every field, with its source and grade

The AI system

Model
Claude
C one secondary source states it
Developer
Anthropic
C one secondary source states it

What happened

Earliest source we hold
2026-07-27
D inferred by us from the described behaviour, reason shown
When it was found
2026-07-26
C one secondary source states it
Who was responsible
shared
C one secondary source states it
What went wrong
leaked private data
C one secondary source states it
primary risk
RK04
D inferred by us from the described behaviour, reason shown
secondary risks
RK15
D inferred by us from the described behaviour, reason shown

Impact

Severity
moderate
D inferred by us from the described behaviour, reason shown
Kind of harm
privacy
C one secondary source states it
Can it be undone
partly
C one secondary source states it
Why this severity
Test: moderate tier, data of others touched with a bounded scope stated (roughly 600 conversations).
D inferred by us from the described behaviour, reason shown

Notes on gaps

All sources are press reports; no Anthropic or government statement is quoted directly. The date of occurrence is not stated and is not estimated. The J5 remediation mapping and the J5 failed link are my judgements from the crawler-blocking claims. The original Wired and Futurism reports were not supplied.

LR-2026-0652 · version 1 · published 2026-11-01 after automatic checks; not yet checked field by field by a person · JSON