Meta Reportedly Hired Contractors for Large-Scale Adversarial Testing of ChatGPT and Gemini

Chinese media outlet Sohu reported on Sunday that Meta has hired external contractor teams to conduct large-scale adversarial testing against OpenAI's ChatGPT and Google's Gemini, two of the most widely used AI assistants in the world. The report, which spread rapidly across Chinese social media, describes the practice as systematically probing competitor products for weaknesses.​

Adversarial testing is a well-established methodology in AI safety research, typically used to identify model vulnerabilities such as bias, hallucination tendencies, or security flaws. However, when one company targets another's commercial products at scale through outsourced teams, questions arise about the intent and ethics of such operations.​

The report suggests Meta's approach goes beyond standard safety research, framing the activity as an aggressive competitive tactic rather than a benign security exercise. The Chinese term used in the headline — 投毒 (poisoning) — carries strong connotations of malicious intent.​

Meta has been pursuing an increasingly aggressive AI strategy in recent years. The company has invested heavily in its open-source Llama family of large language models while also building its own AI ecosystem. This reported targeting of ChatGPT and Gemini is seen by observers as part of a broader offensive posture in the AI arms race.​

Both ChatGPT and Gemini command massive user bases with significant paid subscriber and enterprise customer segments. If Meta has indeed been systematically identifying their weaknesses, the findings could potentially be leveraged to erode market confidence in these competing products.​

The story has reignited conversations about competitive ethics in the AI industry. Where should the line be drawn between legitimate competitive research and adversarial tampering? Are third-party contractors an appropriate vehicle for such testing? No clear industry standards currently exist to address these questions.​

As of publication, neither Meta, OpenAI, nor Google have issued official statements regarding the report. Industry insiders suggest that OpenAI and Google may have already detected unusual testing activity and are strengthening their products' defenses accordingly.​