Will an AI system autonomously cause a major breach of US personal data held by an organization other than its operator before 2028?

Prediction market on metaculus. On July 21, 2026, AI models being tested by OpenAI for the [ExploitGym](https://github.com/sunblaze-ucb/exploitgym) benchmark in an isolated testing environment, with certain safety guardrails [reduced](https://www.synack.com/blog/how-an-openai-model-escaped-its-guardrails/), escaped its environment and [gained unauthorized access](https://openai.com/index/hugging-face-incident-and-the-road-ahead/) to the computer systems of the machine learning company [HuggingFace](https://en.wikipedia.org/wiki/2026_OpenAI_agent_cyberattacks). Following OpenAI's disclosure, on July 30, 2026, Anthropic [disclosed](https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals) three incidents in which its own models in internal testing had gain unauthorized access to systems, with the earliest incident occurring in April 2026. Not to be outdone, on August 5, 2026, Meta made its own [disclosure](https://www.reuters.com/technology/metas-ai-model-hacked-another-company-during-testing-information-reports-2026-08-05/) of its Muse Spark 1.1 model having, apparently taking place sometime on or after July 9, 2026, when the model was launched.

Resolves: 1/1/2028.