Anthropic Added a New Security Measure to Get Back Into the Trump Administration’s Good Graces

The WIRED framing of this as Anthropic 'getting back into good graces' implies the security measure was reactive and politically motivated, not a routine safety improvement. That framing matters because it positions compliance infrastructure as a negotiating chip rather than a principled engineering decision.

The Decoder's coverage from July 1 fills in the technical specifics the WIRED piece leaves vague: the security measure in question is a new safety classifier targeting a jailbreak discovered by Amazon researchers, achieving a 99-plus percent block rate but generating more false positives on benign requests. That cost is real and ongoing for users. The same Decoder piece also reported that the underlying exploit affects smaller models like Haiku 4.5, meaning the fix Anthropic used to satisfy regulators does not close the systemic exposure across its lineup. Separately, the hidden monitoring logic found in Claude Code, also reported by The Decoder on July 1, adds pressure to Anthropic's credibility with enterprise buyers at exactly the moment it needs to rebuild trust with both government and commercial channels.

Watch whether Anthropic publishes a technical disclosure on the classifier's false positive rate within the next 30 days. If it does not, that suggests the compliance measure was sized for regulatory optics rather than production reliability.