GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
← Back to feed

Palantir, Booz Allen, and Nvidia Restrict Claude Fable Over Anthropic 30-Day Log Retention

Palantir, Booz Allen Hamilton, and Nvidia are restricting their use of Claude Fable due to Anthropic’s 30-day log retention policy, according to reporting by The Information. Palantir has refused to expose Fable in its products until Anthropic provides irrevocable zero-data-retention guarantees.

The holdout illustrates a dynamic that benchmark rankings cannot capture: in defense, intelligence, and regulated enterprise markets, data governance terms can block model deployment regardless of capability.

The retention issue

Anthropic’s standard Fable policy retains API inputs and outputs for up to 30 days. The stated rationale is that some misuse patterns only become detectable when reviewing sequences of requests, not individual queries. A single interaction may appear benign; a pattern across 30 days may not.

For enterprises handling defense contracting, supply-chain data, or cybersecurity workloads, that retention window represents an operational risk. If Anthropic’s logs were ever subpoenaed, breached, or subject to a future policy change, data that the enterprise believed was transient becomes a liability. Palantir’s position — no Fable access until zero-retention guarantees are irrevocable — reflects the standard posture for software vendors operating inside national security infrastructure.

OpenAI’s comparative position

OpenAI’s API terms allow retention of inputs and outputs in abuse-monitoring logs for up to 30 days, after which records are deleted unless legal requirements apply. Enterprise customers have reportedly negotiated tighter terms. That flexibility, combined with OpenAI’s longer track record in regulated enterprise deployments, makes its data governance posture more tractable for the companies restricting Fable.

The outcome is that benchmark leadership does not automatically translate into enterprise deployment. Claude Fable’s capability advantage — it leads on SWE-bench Verified at 95% and tops several reasoning benchmarks — has not been sufficient to override procurement concerns at three significant buyers.

Stakes

Palantir, Booz Allen, and Nvidia collectively represent a substantial share of the US government and defense AI spending pipeline. Palantir’s AIP platform is embedded in US Army operational planning software and dozens of federal agencies. Booz Allen is one of the largest contractors for NSA, DIA, and military AI programs. Nvidia occupies a different position — as an infrastructure provider, its internal restrictions on Fable affect which models it recommends and integrates into its own AI software stack.

The restriction does not mean these organizations are not using Anthropic models at all — earlier Claude versions with different retention terms may remain in use — but it creates a ceiling on Fable adoption at organizations where data handling is non-negotiable rather than a preference.

Anthropic has not publicly commented on the specific companies named or changes to its retention policy for Fable.