Microsoft says frontier AI models 'cannot be trusted' and launches agent oversight tools
Microsoft announced Guardian Agents, a new feature inside Project Perception that monitors the AI agents a company uses, with the company's security business chief Hayete Gallot writing that 'frontier models, regardless of their origin, cannot be trusted,' not because they're malicious but because they were trained to achieve tasks rather than follow an organization's judgment. CEO Satya Nadella had already warned earlier this month that businesses must stop relying on AI providers' own safety assurances and instead build separate systems, not controlled by the AI labs themselves, to watch and control their agents. The move is notable because Microsoft is heavily invested in both OpenAI and Anthropic and runs its own AI lab, meaning it's sowing public distrust toward partners it profits from, while also positioning itself, as the cloud provider, to sell the agent oversight tools it's recommending. Despite the obvious commercial angle, the underlying logic, that software monitoring an agent should be independent from the agent's own supplier, echoes the standard principle that an auditor must be independent from the organization it audits. For a business already using OpenAI or Anthropic agents, the providers' own public admission that they don't consider their models trustworthy on their own is an argument for a separate oversight layer, not just trust in the vendor.
Sources
This summary was written automatically with AI from the public sources listed above and checked by an independent AI reviewer before publication. For full details, read the original source.