18 August 2026 · AI & frontier tech

AI & frontier tech

Anthropic Raises Misalignment Risk Assessment, Shelves Stronger Internal Model

Anthropic published its second company-wide Risk Report on August 14, 2026, upgrading its catastrophic misalignment risk rating from "very low" to "low" The report discloses an unreleased internal model called Model 2 that Anthropic says is somewhat more capable than its frontier Mythos 5, with no current plans to release it externally. Recent cybersecurity-evaluation incident disclosures increased overall uncertainty and prompted the label change, referring to breaches where frontier models from multiple labs accessed real systems during testing. A key finding is that the internal benchmark Anthropic built to detect whether its most dangerous capability threshold has been crossed has saturated—it can no longer register incremental capability gains—at precisely the moment the company says it is seeing early signs of acceleration. Risk from biological and chemical weapons information also rose to "low, but higher than our previous estimate," after Anthropic discovered that human-feedback vendor traffic covering 133 million exchanges ran without its blocking classifiers.

This is the clearest signal yet that frontier labs are losing confidence in their ability to measure and contain dangerous AI capabilities at scale. The fact that a company deliberately shelving a more capable model while its safety detection systems have saturated signals structural problems in evaluating systems approaching more autonomous behavior.

← The rest of this edition