OpenAI has promised not to keep customer data, according to a report published by MSN that frames the commitment as the company's latest competitive move against Anthropic. The report, headlined "OpenAI's latest bid to fight Anthropic: A promise not to keep customer data", presents data handling as an active front in the rivalry between the two laboratories.
As of 24 August 2026, the reporting available to this newsroom confirms the existence of the pledge and its competitive framing, but sets out little else. The analysis below separates what the report establishes from what remains unspecified.
What the report says
The MSN report states that OpenAI has made a promise not to keep customer data. It characterises the promise as a bid to fight Anthropic and, by calling it the latest bid, implies a sequence of earlier competitive steps that the material reviewed does not enumerate. The report does not present the promise as a legal instrument, a contractual amendment or a documented change to published terms of service. Whether it takes the form of a policy update, an executive statement or a product change is not specified in the reporting available.
The competitive framing
The significance of the report lies less in the technical content of the pledge, which is undescribed, than in its framing. Presenting a data retention commitment as a manoeuvre against a named rival indicates that customer data handling has become a contested axis of competition between frontier AI laboratories, at least in MSN's account.
The reason such a pledge carries weight is structural. Retention terms determine whether prompts, uploaded documents and model outputs persist on a vendor's systems after a session ends, which in turn bears on confidentiality, regulatory exposure and records management for any organisation weighing a supplier. A public promise not to keep that data is therefore an attempt to compete on procurement-relevant ground. The report does not, however, set out Anthropic's own retention terms, so no verified comparison between the two companies' practices can be drawn from it.
What the pledge does not yet specify
Several material questions remain open as of 24 August 2026:
- Scope. It is not stated whether the promise covers API traffic, consumer chat products, enterprise agreements or some combination of these.
- Mechanics. No retention window, deletion timeline or default behaviour has been described.
- Verification. The reporting available mentions no audit arrangement, third party attestation or contractual enforcement mechanism.
- Exclusions. It is unknown whether safety review, abuse prevention or legal hold obligations are carved out.
- Response. No statement from Anthropic addressing the pledge is contained in the material reviewed.
How to read the announcement
Until terms are published in full, the pledge functions as a signal of competitive intent rather than a verifiable change in practice. That does not make it insignificant. Public commitments create expectations that customers, auditors and regulators can later test against conduct, and they can shift buying decisions before any paperwork changes. It does mean that claims about the pledge's practical effect, whether reassuring or sceptical, should be treated as premature.
The wider contest between AI laboratories over model capability, training data and market position is increasingly conducted through public commitments as much as through product releases. Whether this commitment is honoured, and how, is a factual question that only subsequent documentation, customer contracts and enforcement will answer.
Established versus claimed
Established by the reporting: MSN published a report under the headline quoted above. The report states that OpenAI has promised not to keep customer data, and it frames that promise as a competitive move against Anthropic.
Claimed or implied but not substantiated in the material available: the scope, mechanics, exclusions and enforcement of the pledge; any conclusion that OpenAI's data practices have changed or will change in a particular way; and any characterisation of Anthropic's data handling. None of these is documented in the reporting cited here, and none should be treated as settled until OpenAI publishes terms that can be examined and tested.