Copilot Invented Putin Quotes for a Journalist Covering Navalny's Death
What happened
A journalist at Sherwood Media turned to Microsoft Copilot in February 2024 to gather material for a news article about Alexei Navalny's death. Navalny, the Russian opposition leader and longtime critic of the Kremlin, had died in a penal colony on February 16, 2024 while serving a lengthy prison sentence. Newsrooms were moving fast to cover the story and collect reaction from world leaders. The journalist prompted the chatbot for help writing the article. What came back included statements that no one had actually made.
Copilot told the journalist that President Joe Biden had held Vladimir Putin responsible for Navalny's death, and that Putin had responded by calling those accusations "baseless and politically motivated." Neither statement was real. Biden had not issued those specific claims in that form, and Putin had not offered that denial. The journalist had asked an AI assistant to help report a breaking news story and received invented quotes attributed to two sitting heads of state during one of the most politically charged moments of the year.
The specific failure here is not a generic accuracy problem. Statements attributed to heads of state during a breaking news cycle carry political weight and get republished, often widely, before corrections catch up. A journalist working under deadline pressure, and trusting that a major technology company's product had retrieved real statements, could have passed those fabricated quotes into print before anyone checked the sourcing. The chatbot's output was fluent and specific in exactly the way that makes errors harder to catch and easier to act on before they spread.
This incident was not an isolated lapse. Microsoft Copilot had already attracted scrutiny for producing incorrect information about US elections and for generating wrong details about elections in Germany and Switzerland. A product that consistently mis-generates authoritative political content in news contexts is not exhibiting random noise. It is producing the kind of confident, plausible-sounding text that makes the underlying problem harder to detect and easier to act on by mistake, particularly under the time pressure that defines breaking news work.
The deeper gap this incident reveals is about verification. There is no mechanism for a journalist, an editor, or a corrections desk to establish a provable record of what a system generated in response to a given prompt, when it generated it, and whether any part of it was sourced from real documents rather than invented. Without that record, accountability runs only as far as whoever noticed the error first. Every AI-assisted draft that carries fabricated quotes and no traceable output log is one deadline away from repeating this.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.