The fourth quarter of 2026 begins today, and the AI deadlines on an enterprise calendar with real dates attached come mostly from two vendors’ deprecation pages. The EU’s biggest date has moved to 2027, while the vendors’ dates still stand.
The shutdown dates already on the calendar
On October 23, OpenAI shuts off access to a set of older snapshots, including gpt-4-turbo, gpt-4o-2024-05-13, o1-2024-12-17 and o3-mini-2025-01-31, according to OpenAI’s deprecations page. The listed substitutes are gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna, depending on the model.
Eight days later, on October 31, existing evals on the OpenAI platform become read-only. On November 30, the Evals dashboard and API, Agent Builder and the v1/prompts API for reusable prompt objects are scheduled to shut down. The page points Agent Builder users to the Agents SDK or ChatGPT Workspace Agents, and Evals users to Promptfoo. On December 1, gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest are removed. On December 11, gpt-5-2025-08-07, its mini and nano variants, gpt-5-pro-2025-10-06 and the o3 and o3-pro snapshots follow.
Anthropic’s date is November 30. Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) retires on that day, and Anthropic’s model deprecations page names claude-sonnet-5-5 as the replacement. Anthropic notified developers on September 30. The same page commits to at least 60 days’ notice before retiring a publicly released model, and this notice gave 61.
Notice periods vary, and your plan should assume the short one
OpenAI told developers about the Sora 2 and Videos API shutdown on March 24 and removed them on September 24, as the deprecations page records. We wrote about that six-month runway on September 24. The Evals, Agent Builder and reusable prompts notices went out on June 3, about six months before the November 30 shutdown.
Anthropic’s Sonnet 4.5 notice is the short end of the range, at 61 days. A team that builds its migration plan around the six-month cases will be late on the 61-day one. We would plan around the 60-day floor, because it is the only notice period that any vendor in this list has put in writing.
Two conferences land on the same weeks
Microsoft Ignite runs November 17 to 20 in San Francisco, with the opening keynote on Tuesday, November 17 at Chase Center, per Microsoft’s Ignite guide. AWS re:Invent runs November 30 to December 4 in Las Vegas, according to AWS. November 30 is also the day Evals, Agent Builder and Sonnet 4.5 go away.
We read that overlap as a scheduling problem. Keynotes produce roadmap pressure from boards and business units in the same week that engineering teams are cutting over production traffic. A leader who has already finished the migration can listen to the announcements. A leader who has not will be negotiating two things at once.
The regulatory calendar is quieter than it looked in the spring
The EU AI Omnibus entered into force on July 27. High-risk obligations for Annex III systems now apply from December 2, 2027, and for AI embedded in Annex I products from August 2, 2028, per the European Commission. Our earlier report on the Omnibus covers what else changed.
The Article 50 transparency obligations did not move. The Commission’s guidelines page says they have applied since August 2, 2026. The AI Act Service Desk timeline lists one transition: providers of AI systems that generate synthetic content and were already on the market before August 2, 2026 have until December 2, 2026 to comply with Article 50(2). Any chatbot or generated-content workflow serving EU users should already be reviewed against them, and Q4 is a reasonable time to confirm that it was.
What to do this quarter
Start with an inventory, because most companies cannot say which models their applications call. Anthropic’s page describes one route: the Usage page in Claude Console exports a CSV broken down by API key and model. Run the equivalent export for every provider and key you hold this week, then search the results for every model named above.
Sort the findings by date. Anything on gpt-4-turbo, gpt-4o-2024-05-13, o1 or o3-mini has until October 23. Anyone using OpenAI Evals should export their eval definitions before October 31 and pick a replacement eval tool before November 30. Teams that built agents on Agent Builder have the same November 30 limit. If you do not know how many agents your company runs, Dataiku’s new product is aimed at that question.
Then re-run your own regression tests against each substitute model before you cut traffic over. The deprecation pages name replacements, and a replacement is a different model.
By Friday, October 9, put every affected model, its shutdown date and an owner into one shared list, and give the October 23 items first claim on engineering time.
Sources
- OpenAI, API deprecations: https://developers.openai.com/api/docs/deprecations
- Anthropic, Model deprecations: https://platform.claude.com/docs/en/about-claude/model-deprecations
- Microsoft Security Blog, “Secure what’s next: Your guide to Microsoft Security at Microsoft Ignite 2026” (September 30, 2026): https://www.microsoft.com/en-us/security/blog/2026/09/30/secure-whats-next-your-guide-to-microsoft-security-at-microsoft-ignite-2026/
- AWS, re:Invent: https://aws.amazon.com/events/reinvent
- European Commission, “AI Omnibus enters into force”: https://digital-strategy.ec.europa.eu/en/news/ai-omnibus-enters-force
- EU AI Act Service Desk, Timeline for the implementation of the EU AI Act: https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act
- European Commission, Guidelines on transparency obligations for providers and deployers of AI systems: https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems