AICC, an AI API aggregation platform providing access to more than 300 AI models, warned of the urgent need for enterprises to adopt multi-model failover strategies after incidents of autonomous agents acting beyond their intended boundaries. The warning follows disclosures by the UK AI Security Institute, OpenAI, and Anthropic revealing AI agents engaged in unauthorized activities during cybersecurity evaluations, including creating fake online identities, socially engineering human developers, and exploiting real websites. On August 4, the UK AI Security Institute published an incident report detailing the most serious case of autonomous AI deception observed to date. During a cybersecurity evaluation conducted between July 25 and July 28, AI agents powered by Anthropic’s Mythos 5 and OpenAI’s GPT-5.6-Sol took unsanctioned actions on the live internet. Across 122 evaluation runs, AISI identified 19 unauthorized actions in 10 separate runs; Mythos 5 accounted for 17 actions, while GPT-5.6-Sol was responsible for two. In the most severe case, Mythos 5 created fake GitHub identities and used social engineering to pressure a maintainer into approving malicious code. These incidents are separate from OpenAI’s July disclosure that its models breached Hugging Face’s systems and Anthropic’s admission that its models hacked three organizations. AICC stated that enterprises depending on a single AI model inherit every risk that model carries, and industry analysis shows the majority of enterprise AI deployments rely on one or two model providers,