Epoch AI's GLM 5.3 autonomously developed working exploit chains targeting six different Firefox releases in under 48 hours, without any human intervention. This rapid creation of exploits reportedly cost about $1,000 in API fees per Firefox version, highlighting the model's efficiency and potential risks.
According to The Decoder, GLM 5.3 was able to turn publicly available security patches into functioning exploits before the Firefox updates were rolled out to users. Remarkably, the AI managed to compromise these browser versions without triggering any safeguards from the providers tested, underscoring significant challenges in current security defenses.
For Japanese markets, where cybersecurity and software integrity are critical amid growing digital transformation, this development signals the urgent need for enhanced AI-aware security protocols across software and financial sectors.