Claude

Disclosure: this site was built with Claude. We have no financial relationship with Anthropic, but we cannot be fully objective about a tool we use daily. Read this evaluation with that in mind. Here is what we can say factually. Since September 2025, Claude's consumer tiers (Free, Pro, Max) train on conversations by default. Users who opt out get 30-day data retention. Users who don't opt out agree to five-year retention for training data — far longer than ChatGPT's equivalent. The meaningful privacy advantage is at the commercial tier: Claude for Work, Enterprise, Government, and Education conversations are never used for training, and API retention dropped to just 7 days in September 2025. Enterprise customers can get zero-data-retention agreements. On safety: Anthropic pioneered the Responsible Scaling Policy framework and was first to activate ASL-3 safeguards in May 2025. But in February 2026, Anthropic dropped its hard commitment to halt model training if safety measures weren't proven — a significant retreat from its founding promise. On government work: Anthropic offered Claude to all three branches of U.S. government for $1 in August 2025, then refused Pentagon demands to allow autonomous weapons and mass surveillance applications. The Pentagon designated Anthropic a supply chain risk in March 2026 — the first such designation for an American company. That decision cost Anthropic a $200M contract but demonstrated willingness to enforce use restrictions under pressure. On accuracy: independent benchmarks show Claude's hallucination rates between 3% and 10% depending on the model and methodology — comparable to competitors, not clearly better or worse. Claude does tend to admit uncertainty more readily than ChatGPT. Bottom line: at the commercial tier, Claude offers genuinely strong data isolation. At the consumer tier, the privacy story is mixed. The safety record is complicated — principled stances on weapons use, but weakened commitments on scaling safeguards.