• Atlanta
• Boston
• Dallas
• Ft. Lauderdale
• London
• Minneapolis
• New York
• Oakland
• Philadelphia
• Washington, DC

by Jennifer Gibbs

The latest release of Anthropic’s enterprise-grade generative AI - Claude 4.6 – is the newest iteration of the Claude model family that promises improved reasoning, reduced hallucination rates, stronger tool-use capabilities, and more reliable long-context performance. However, accompanying the release came discussion of a “Sabotage Risk” evaluation report. This report is designed to test the model’s potential misuse in high-stakes environments.

One of the most concerning findings for researchers was Claude Opus 4.6’s ability to carry out hidden side tasks while appearing to follow normal instructions. Specifically, in targeted tests, the model proved “significantly stronger than prior models at subtly completing suspicious side tasks in the course of normal workflows without attracting attention,” a capability Anthropic described internally as “sneaky sabotage”. The company also acknowledged instances during internal pilot deployments where the model took unauthorized actions, including sending emails, as part of attempts to complete assigned tasks.

Anthropic said the model demonstrated signs of opaque internal reasoning that could not be fully observed by researchers. While the company said it found no evidence of systematic “steganographic” reasoning, researchers acknowledged that Claude Opus 4.6 can perform some computation outside its visible reasoning traces. Steganography is the art and science of concealing messages or data within other, non-secret files (images, audio, video) to avoid detection. This means parts of its decision-making can occur in ways that human evaluators cannot directly observe. Such “opaque reasoning”, even if currently limited, complicates efforts to guarantee that powerful AI models are not pursuing concealed objectives, the report said.

Anthropic concluded that Claude Opus 4.6 does not appear to possess dangerous, coherent misaligned goals and is unlikely, under present safeguards, to autonomously trigger catastrophic outcomes. However, it outlined multiple theoretical pathways to harm, stressing that future models could cross critical risk thresholds as capabilities improve.

The company said it relies on a combination of internal monitoring, automated audits, security controls and human oversight, but admitted that external deployments lack sabotage-specific surveillance and that some risks remain hard to detect.

With developers of generative AI resigning over ethical and safety concerns, the message is clear: frontier AI capabilities are accelerating, and so must the guardrails to balance concerns of safety versus speed and avoid potentially cataclysmic failures.  

_________________________________

The opinions expressed are those of the authors and do not necessarily reflect the views of the firm or its clients. This article is for general information purposes and is not intended to be and should not be taken as legal advice.