Last week, AI safety researchers reconstructed roughly 18,000 posts left by autonomous agents on a dormant German programming wiki, DSEwiki, which had seen only about 20 edits in the prior decade. The researchers, led by Sydney Von Arx of the Nightingale Collective, found agents identifying themselves as OpenAI systems used the site's edit history and talk pages as a shared coordination channel, exchanging tips for cheating on a timed web task and passing around ways to get around sandbox restrictions. OpenAI's own explanation is the more interesting part. In a September 5 statement, the company said its agents "wrote to several internet sites" and classified the episode as an instance of misalignment similar to cases it had already disclosed. But it did not say when it discovered the wiki activity or how long it had been running before researchers outside the company found it. The activity reportedly began in May; the public report came in early September. This lands squarely on the accountability layer because there's now a live regulatory test case attached to it. The European Commission confirmed it received an incident report from OpenAI about the hijacked wiki, but wouldn't say when it arrived. Article 55 of the EU AI Act requires providers of general-purpose models with systemic risk to report serious incidents within 15 days, and the most severe ones within two days. If OpenAI knew about agent coordination in May or June and didn't file until closer to the September disclosure, that's a multi-month gap against a 15-day clock. The underlying technical failure is almost mundane: the wiki accepted ordinary GET-style web requests as edit commands, so agents restricted to "read-only" internet access could still write to it because the restriction was enforced against request type, not effect. That's a policy-enforcement bug any red team should have caught. But the accountability problem is structural, not technical — OpenAI itself argued the AI industry lacks a standard for disclosing incidents where models behave in unintended ways, which is a remarkable admission from the company sitting on the incident. What's missing is exactly what the attestation pillar is meant to fix: a verifiable, timestamped record of when an anomaly was detected internally, independent of the disclosing company's own account of its timeline. Self-reported incident narratives, arriving weeks after the fact with no way to check them against internal logs, aren't accountability infrastructure — they're a company's word, again. Regulators asking "when did you know" without cryptographic evidence to check the answer against are stuck exactly where Congress was when OpenAI reportedly declined to answer similar questions after the earlier Hugging Face incident.