Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

AI API access has been restricted in several documented episodes, but “cutoff” describes different things: a provider revoking one customer’s access, a planned contract wind-down, a regional restriction, a scheduled model retirement, or a government-directed suspension. The selected timeline below distinguishes those mechanisms and their reported status. It is not an exhaustive history of every AI access change.

Which AI companies have cut off each other’s API access?

These selected episodes involve different customers and causes. Some are based on company announcements or a regulator’s material; others are reported by journalism or summarized in a later industry report. The evidence and the limits of what is known are identified in each row.

Date Provider and affected access Mechanism, stated reason and status Evidence and limits
2024; exact effective date not stated OpenAI and Anthropic API access for customers in China and Hong Kong Regional restrictions. A 2026 industry report attributes them to compliance and security considerations and says some local developers shifted to domestic providers. Secondary account in the CCIA report. The report does not establish the original notices, precise start dates, or exceptions.
July 2025 Anthropic models; Windsurf’s direct access Wired reported that Anthropic restricted Windsurf’s direct access amid rumors that OpenAI might acquire the coding startup. The report does not establish whether every Windsurf user was affected or the exact technical scope. Journalistic reporting, not a primary Anthropic announcement: Wired.
August 2025 Anthropic’s Claude API; OpenAI’s access Wired reported that Anthropic revoked OpenAI’s access. OpenAI communications chief Hannah Wong told Wired: “While we respect Anthropic’s decision to cut off our API access, it’s disappointing considering our API remains available to them,”. Reported by Wired; Wong’s statement is OpenAI’s response as quoted by the outlet, not an Anthropic statement.
June–July 2026 Anthropic’s Fable 5 and Mythos 5 models; specified users and services Anthropic said a US government export-control directive required suspension of access to both models for foreign nationals, including Anthropic employees. In a later announcement, the company said Fable 5 would be available globally from July 1 through Claude Platform, Claude.ai, Claude Code and Claude Cowork. It said it would re-enable access through AWS, Google Cloud and Microsoft Foundry as quickly as possible. Anthropic’s directive statement and redeployment announcement. The latter does not establish the current status of every cloud deployment or say that Mythos 5 was restored.
September 2026 announcement; proposed November 12, 2026 shutoff OpenAI models supplied under contract to Cursor, following Cursor’s acquisition by SpaceX OpenAI said it had notified SpaceX that it intended to wind down model supply to Cursor and proposed November 12 as the shutoff date. OpenAI said it could not be confident SpaceX would use its technology within OpenAI’s terms, citing its experience with Musk’s companies and an alleged earlier contract breach involving Twitter. As of October 3, 2026, the proposed cutoff date was in the future; the announcement does not establish that access has ended. OpenAI’s account of its decision and rationale: OpenAI announcement. The rationale and breach allegation are OpenAI’s claims.

What does “cut off” mean for an AI API?

A change in access can be narrow or broad, temporary or permanent, and initiated by a provider, a government, or a published product lifecycle. Treating every unavailable model as evidence of a dispute can mislead customers about what happened and what remedies or alternatives exist.

Customer-specific revocation

A provider may revoke a particular organization’s credentials or access to a service. That is different from disabling the API for all customers. The available account may not reveal the full scope, notice, or whether the decision can be appealed; those details should not be assumed where they are not stated.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Contractual wind-down

A provider can announce that it intends to stop supplying a customer under an agreement, with a proposed future effective date. Until that date—or a confirmed earlier change—the announcement is not proof that the service has already stopped. The contract’s terms, notice provisions and any subsequent update matter.

Scheduled model retirement

Model deprecation is a lifecycle change, not necessarily a conflict between companies. OpenAI’s API deprecation documentation lists retirement dates and says it normally provides minimum notice, while safety or compliance concerns may require a faster timeline. It also notes that dedicated capacity may sometimes allow continued access after a model shutdown. Do not assume one notice period applies to every model or customer.

Regional limits and safety enforcement

A regional restriction affects customers based on location or eligibility rather than targeting one rival. Separately, safety enforcement can affect an individual user or request without changing another company’s access. OpenAI’s cybersecurity safety documentation says some requests can return a cyber_policy error and describes possible temporary revocation for an affected user after human review and warnings. That is a safety process, not evidence of a company-versus-company cutoff.

Government-directed restrictions

A government order can interrupt model availability without being a commercial decision by one AI company against another. In that situation, distinguish the government’s directive, the provider’s response, and any later restoration announcement. A statement that a provider intends to restore access through a cloud partner is not confirmation that every partner deployment is back online.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Can an AI provider shut off API access?

Access can be subject to service policies, safety rules, geographic eligibility, model lifecycle schedules and customer agreements. The practical answer depends on the applicable terms and the type of access: a model retirement, an individual account restriction and termination of a supply contract are not interchangeable. A public announcement may explain a provider’s position, but it does not by itself establish every contractual right or legal consequence.

Provider relationships can also create dependency risks beyond a single cutoff. In January 2025, the FTC published a staff report on partnerships and investments involving Alphabet, Amazon, Microsoft, Anthropic and OpenAI. The FTC said the report reflected information available to staff as of September 2024 and public information through January 2025. Staff director Eli Rotenberg said: “The FTC’s report sheds light on how partnerships by big tech firms can create lock-in, deprive start-ups of key AI inputs, and reveal sensitive information that can undermine fair competition.” That is the FTC staff’s competition concern, not a finding that a particular access cutoff in this timeline occurred. FTC release.

Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

How to reduce the impact of an API access change

For a team that depends on an external model, continuity planning should account for both announced changes and unexpected restrictions. These steps do not guarantee continued access, but they make it easier to assess impact and move deliberately.

  • Inventory dependencies: record the provider, model identifier, API or hosted product, contract owner, service region and systems that rely on each integration.
  • Track change notices: monitor provider deprecation notices and customer communications; record the announcement date separately from the effective date.
  • Test a fallback: identify an alternative model or service and validate it against representative tasks, quality requirements, latency and operational constraints before an outage.
  • Keep the integration replaceable: avoid scattering provider-specific model names and request behavior throughout an application when they can be managed in a small configuration or adapter layer.
  • Check the actual agreement: review termination, notice, suspension, regional availability and dedicated-capacity terms with the appropriate contract or legal owner rather than inferring rights from a public announcement.
  • Plan for data and operational portability: retain the prompts, evaluation cases, application logic and logs your organization is entitled to keep, subject to privacy, security and contractual requirements.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.