← All ridiculous benchmarks

Wiretap Bench

Wiretap BenchOpenAI: 1. OpenAI reported finding a hidden recording device in an office, using its models to investigate the device and its origin, and referring the matter to the FBI.; Anthropic: 0. No matching FBI referral in this source record; not a model evaluation.; Google: 0. No matching FBI referral in this source record; not a model evaluation.; Meta: 0. No matching FBI referral in this source record; not a model evaluation.Wiretap BenchFinally, a benchmark that comes with a case number.FBI called00.51OpenAI: 1 FBI called. OpenAI reported finding a hidden recording device in an office, using its models to investigate the device and its origin, and referring the matter to the FBI.1OpenAIAnthropic: 0 FBI called. No matching FBI referral in this source record; not a model evaluation.0AnthropicGoogle: 0 FBI called. No matching FBI referral in this source record; not a model evaluation.0GoogleMeta: 0 FBI called. No matching FBI referral in this source record; not a model evaluation.0MetaCompanyOpenAI Newsroom · Oct 13, 2026. Zero = no referral in this record.franklineh.com/ridiclousbenchmark

Swipe the chart to see the full comparison.

Recorded referrals as of . Zero means no matching referral in this source record.

Wiretap Bench gives one point for a reported FBI referral involving a hidden recording device at an AI company’s office. OpenAI takes the lead with one. Anthropic, Google, and Meta have zero entries in this source record. The sample size is doing its best.

On October 13, 2026, OpenAI said it found a recording device hidden in one of its offices. It reported using its models to understand the device and trace its origin, then referring the matter to the FBI. That announcement inspired Franklin’s original Wiretap Bench chart.

How the score works

One reported referral earns one point. We count the event once, regardless of how many models helped investigate it or how many calls someone made. The announcement does not name a model, so the bars compare companies.

A zero means this record contains no matching referral for that company. It does not mean we tested its models, searched every office, or established that no incident occurred. These are the same four companies and scores shown in the original chart.

The source behind the winning bar

OpenAI’s announcement supports the single point. Its account describes the discovery and investigation; Wiretap Bench makes no claim that a model found the device on its own. The source links let you read the announcement and see the chart that started the joke.

Explore the other ridiculous benchmarks or compare model performance in our AI benchmarks.