Now loading...
The head of one of England’s biggest police departments has acknowledged that an artificial intelligence tool from Microsoft fabricated details about a soccer game, which then appeared in an official security assessment. This error contributed to restrictions on Israeli supporters attending a European soccer fixture last year.
Craig Guildford, the chief constable of West Midlands Police, disclosed in a recent submission to lawmakers that the AI system, known as Copilot, generated false information about a supposed match between West Ham United and Maccabi Tel Aviv. The invented result was incorporated into a police intelligence document without verification, Guildford explained in his letter to the Home Affairs Committee.
The flawed report influenced the Birmingham Safety Advisory Group’s classification of a Europa League contest between Aston Villa and Maccabi Tel Aviv as a high-risk event in November 2024. As a result, fans of the Israeli club were prohibited from the game, citing concerns over potential violence stemming from earlier incidents involving Maccabi supporters in Amsterdam, including fights and hate crimes. Authorities highlighted these prior disturbances as key factors.
Guildford’s admission marks a shift from his earlier accounts. In December 2024 testimony before the committee, he rejected any involvement of AI in compiling the report, attributing the inaccuracy to data pulled from social media. Just weeks ago, he pointed to a faulty online search result as the culprit. The latest clarification attributes the issue squarely to Copilot, which Guildford noted was used on a Friday afternoon. The committee received the letter this week.
Microsoft includes a cautionary note with Copilot, advising users that the tool can produce inaccuracies. Independent evaluations of the AI have similarly uncovered instances where it misinterprets information or invents details. Efforts to obtain comment from Microsoft regarding this specific fabrication went unanswered.
The incident underscores growing challenges in relying on generative AI for sensitive tasks like public safety planning, especially as law enforcement agencies experiment with such technologies.
