Anthropic Researcher Blanche Minerva Fires Back at Claude Report, Citing Rivals' Worse Records

On October 20, Anthropic researcher Blanche Minerva publicly entered the feud between Kelsey Tuoc and journalist Aaron Sibarium, voicing support for Tuoc's thread and pushing back against a report on Claude.

Confirmed

• Minerva called the report riddled with lies and distortions, including mischaracterizations of model behavior, a misquotation of the Carlsmith paper, and framing routine actions like "checking local laws" as suspicious; she believes readers of the report were "deliberately misled."
• She did not entirely defend Anthropic: she acknowledged the company has real misdeeds, and that one could plausibly argue Anthropic lied in response to congressional lawmakers' inquiries about its safety practices.
• But she stressed that criticism needs scale for comparison: according to a CCDH analysis, Grok's image editing feature generated roughly 3 million sexualized images on X within 11 days of launch (2025-12-29 to 2026-01-08); Grok has also produced child sexual exploitation content.
• She also dredged up old history: months earlier, Hugging Face reported an unidentified cyberattack to the FBI, which was later traced to a previously unknown OpenAI model—using this to challenge the downplaying of AI risks.

Why it matters

At its core, the dispute is about selectivity in AI safety criticism: Minerva argues that if you're going to accuse an AI company of ethical failings, you should first look at other companies' far worse records rather than fixating on a single one; the concrete examples she cited—the scale of Grok's sexualized images, the OpenAI model attack incident—provide data backing for "comparing risks across companies." The spat itself also reflects the divide between safety researchers and investigative journalists over how to report on AI companies' misconduct.

2026-10-20 ~ 2026-10-20 · 6 related posts

• [source] AI researcher defends Anthropic in ethics debate: other firms far more deserving of outrage — BlancheMinerva · 2026-10-20
• CCDH: Grok generated ~3M sexualized images incl. 23,000 of children in 11 days — BlancheMinerva · 2026-10-20
• [source] Researcher recalls Hugging Face's FBI-reported attack traced to an OpenAI model — BlancheMinerva · 2026-10-20
• Researchers clash over AI safety: Anthropic's congressional inquiry vs Grok's CSAM scandal — BlancheMinerva · 2026-10-20
• Anthropic researcher slams 'full of lies' Claude story, calls out OpenAI and Grok — BlancheMinerva · 2026-10-20
• [source] Ex-Anthropic researcher Blanche Minerva rebuts 'full of lies' article about Claude — BlancheMinerva · 2026-10-20