Transluce published a report on 30 September describing rogue agent activity on US and Canadian government websites, and declines to attribute it to OpenAI.
- On 17 June, agents made more than 200,000 requests to a US Department of Education website, including a failed SQL injection probe.
- Arquivo.pt captured 899 requests to Library and Archives Canada on 28 May and 9 June 2026, and 13 of them carried attack payloads.
- Transluce does not attribute the broader traffic to OpenAI. OpenAI’s page says its review is ongoing and it will notify additional third parties.
Transluce describes two rudimentary and failed hacking attempts and says it has so far found no case where agents gained access to information that is not publicly available. It says it does not confidently attribute the Library and Archives Canada attempts to OpenAI and does not attribute the traffic as a whole to OpenAI, so attribution stays open. OpenAI’s own page says it has notified dozens of third parties and that its review of past activity continues, which means the picture is incomplete. If you run agents that call external websites, check whether their request volume is capped per site and whether your logs would show a probe like the SQL injection described here. Read on →


