Wrapping Up the Apart Research AI Incident Response Sprint

From 11 to 13 September 2026, we spent three days working together on an AI safety incident. We hosted the Türkiye leg of the AI Incident Response Sprint organised by Apart and the Centre pour la Sécurité de l’IA - CeSIA.

Over the three days, participants examined a real incident in which AI systems behaved unexpectedly, approaching it from different perspectives. Through AI Safety Türkiye, 51 people reported that they had registered for the Sprint. Sixteen teams were formed, while some participants chose to work individually. On Saturday, 12 September, we also had the opportunity to meet and work together in person in Ankara and Istanbul. Globally, 943 people registered for the Sprint.

Thank you to everyone who dedicated a weekend to this work. At the end of the event, we announced that the projects would be submitted to Apart Research for evaluation and that we would share the results.

Sprint format

The sprint focused on research into understanding, containing and preventing incidents involving unauthorised or unexpected actions by AI systems. Participants could work individually or in teams of up to five, and fully online participation was possible.

The aim was to produce a short written research output over the weekend. Formats could include a checklist, draft standard, policy document or communication kit. Building software was not required: alongside technical projects, there was room for contributions from law, policy, communication and the social sciences.

Programme

The preparation and work schedule we announced before the event was:

• 24 August–9 September: Individual applications, including track preferences and team status.
• 31 August–9 September: Reviewing the resource list and preparing for the tracks. A WhatsApp group was planned for coordination and sharing materials with participants in Türkiye.
• 10 September: Finalising teams and tracks, matching individual applicants and an evening preparation briefing.
• 11 September: The online global kickoff and the start of the sprint.
• 12 September: Continued project work, with optional in-person meetups in Ankara and Istanbul. These meetups gave us the opportunity to work together.
• 13 September: Completing projects and submitting them online.
• After the sprint: Evaluation by Apart Research and the announcement of results.

Five tracks

The programme offered five tracks through which participants could contribute from different perspectives. The following describes their aims and suggested outputs:

• Containment: Designing auditable safeguards for high-risk AI tests, including rules, required evidence and implementation costs.
• Incident analysis — What Happened / What Breaks Next: Examining public incident records to identify detection gaps and prepare checklists other organisations could use.
• Regulatory Response: Defining questions regulators could ask and what would count as an adequate answer; evaluating incident reporting forms using public information.
• Communication: Studying media coverage of the incident and preparing crisis communication kits, journalist FAQs and plain-language explainers.
• Open: Developing projects within the sprint’s scope that did not fit the other four tracks.

The event announcement listed $2,000 in total prizes across all tracks and a fast-track route into the Apart Fellowship for top teams to continue their research with mentorship.

For programme details, see Apart Research’s sprint page.

There are still very few people in Türkiye thinking and talking about AI safety. Would you like to help close this gap?