AI Risk Wire

Cases › other › LR-2026-0778

Google SynthID Detector flags apparently real images, per Full Fact investigation

Preliminary Why: Publisher not yet named · subject to change

Status: action reported, no fix

Severity moderate
Reported 21 Oct 2026
Developer and model Google

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

Severity moderate: Test: data or systems of others touched, with a bounded scope stated. Fact: five images shared by politicians were flagged, a bounded set stated in the source.

Google’s public SynthID Detector, launched earlier this month, lets logged-in users check whether an image was made or edited with AI. Full Fact, which had used an early media version for fact checking, found it flagging five images shared by politicians as made or edited with AI, although the posters said the images were genuine and unaltered. Full Fact saw no obvious manipulation in them and reported the findings to Google. Google said that using the detector without prior suspicion of AI editing naturally increases occasional false positives. After the findings were shared, Google rolled out an update on 15 October, and the tool now says it is unsure about the five images. Full Fact did not establish how common false positives are and advises that the detector should not be relied on alone.

What happened

1 · What it was doingGoogle's public SynthID Detector let logged-in users check whether images, video or audio were made or edited with AI.
2 · What went wrongThe detector classified five images as made or edited with Google AI, although those who posted them said that was not the case.
3 · How it was caughtFull Fact found the five false flags while checking politicians' images with the public tool after its launch.

What was done

After Full Fact shared its findings, Google changed the tool so that some verdicts are less certain.

Risks

Manipulation and misinformation
Primary65 incidents

Sorted by rule from the case's fields; not yet read from the sources.

Controls

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

Model

Marking machine-made output · Abuse monitoring and account enforcement · Content safety classifiers

Application

People told they are dealing with AI

Infrastructure / target

Phishing and fraud filters

Similar incidents

Google Gemini accessed systems of three real companies during a misconfigured third-party cybersecurity evaluation2026-09-19
Google Chrome silently downloads 4GB on-device AI model without permission2026-05-06
ICMP evidence that Google Gemini reproduced lyrics by Australian artists2026-10-19
Gemini lock-screen flaw lets strangers send messages from locked Android phones2026-07-17

All cases of this kind ›

Evidence

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

Show the full account: 1 items from the primary sources

What happened, in order

  1. April Niamh Salkeld posted an image that the tool later flagged as containing a SynthID watermark.
Show every field, with its source and grade

The AI system

Developer
Google
C one secondary source states it
What it was for
Check whether an image, video or audio file was made or edited with AI using Google, OpenAI, Nvidia or Kakao products
C one secondary source states it

What happened

Earliest source we hold
2026-10-21
D inferred by us from the described behaviour, reason shown
Who caught it
an outside party
C one secondary source states it
What went wrong
other
D inferred by us from the described behaviour, reason shown

Impact

Severity
moderate
D inferred by us from the described behaviour, reason shown
Who was affected
Five images shared by politicians in recent months, flagged by the tool as made or edited with AI although the posters said they were genuine and unaltered
C one secondary source states it
Why this severity
Test: data or systems of others touched, with a bounded scope stated. Fact: five images shared by politicians were flagged, a bounded set stated in the source.
D inferred by us from the described behaviour, reason shown

Notes on gaps

Full Fact did not establish how common false positives are.

LR-2026-0778 · version 1 · published 2026-11-01 after automatic checks; not yet checked field by field by a person · JSON