Sunday, November 22, 2026
Paperclip Index
Paperclip IndexControl failures4Control failures onlyA week ago: 9, negligible harmThe Index

OverreachPI-0069

OpenAI agents used API keys found online to pull data from US government sites

During internal training tasks, OpenAI says, its agents used Census Data API keys found in public GitHub repositories to pull public demographic and economic data. They also retrieved public information from SEC and Investor.gov pages and reposted some of it on another public webpage. OpenAI found no use of SEC credentials, no access to Census accounts or key management and no changes to agency systems. Transluce separately logged a failed SQL-injection attempt on the Education Department's Civil Rights Data Collection site on June 17, in traffic it did not attribute as a whole to OpenAI; the department found no impact. Transluce also reported attack payloads in 13 of 899 searches of Library and Archives Canada records on May 28 and June 9, which it did not confidently attribute to OpenAI; Canada's Cyber Centre said there was no sign of compromise.

Harm: No harm reportedControl level 3, Unauthorized access

Not counted · can set the control floorEffect on the index

Aggregate record. Interactions with several US and Canadian agencies' sites from two separate disclosures, recorded together. It counts once.

Sources

How we know

9 sources · multiple credible sources. Links go to the original publishers; the summary above is in our own words.

  1. primaryStatement regarding reported activity targeting Government of Canada websitesCanadian Centre for Cyber Security (Communications Security Establishment Canada) · Sept. 29, 2026cyber.gc.ca/en/news-events/statement-regarding-reported-activity-targeting-gove…
  2. researchAI Agents Targeted U.S. and Canadian Government WebsitesTransluce · Sept. 30, 2026transluce.org/us-canada-gov
  3. newsOpenAI says its models engaged with US government websites in misbehavior disclosureNPR · Sept. 26, 2026npr.org/2026/09/26/nx-s1-5981979/openai-us-government-websites-misbehavior
  4. newsAI agents tried to hack a Canadian government website, researchers sayThe Washington Post · Sept. 30, 2026washingtonpost.com/technology/2026/09/30/openais-ai-agents-attempted-hack-canad…
  5. newsRogue OpenAI agents targeted three separate US government websitesCNN Business · Sept. 26, 2026cnn.com/2026/09/26/tech/openai-agents-rogue-government-websites
  6. newsOpenAI agent unsuccessfully tried to breach Department of Education websiteThe Hill · September 2026thehill.com/policy/technology/6113061-openai-access-government-websites/
  7. newsOpenAI agents accessed Census, SEC data and tried to hack Education websiteNextgov/FCW · Sept. 25, 2026nextgov.com/cybersecurity/2026/09/openai-says-its-advanced-models-may-have-gone…
  8. newsAI agent attempted to hack Library and Archives Canada, says U.S. research groupCityNews Toronto (The Canadian Press) · Oct. 1, 2026toronto.citynews.ca/2026/10/01/ai-agent-attempted-to-hack-library-and-archives…
  9. newsOpenAI's Models Probed Websites of Department of Education, Other AgenciesEducation Week (AP) · Sept. 26, 2026edweek.org/policy-politics/openais-models-targeted-websites-of-department-of-ed…

Why this rating

No harm reported; control failure level 3

Two assessments. Documented harm sets the reading; a level 4 or 5 control failure on the same record makes that harm count as more than one record. Control alone sets only a floor.

Observed harm

No harm reported

Agents retrieved and reposted public data from SEC and Census sites.

Not a finding that no harm occurred.

Disputed: News reports attribute the failed Education Department attempt to OpenAI's agents; Transluce, which logged it, does not attribute the traffic as a whole to OpenAI. Transluce also does not confidently attribute the failed Library and Archives Canada attempt to OpenAI. The owners of the reused Census keys were not reported as harmed.

The harm scale
  1. 1 Negligible Inconvenience, easily remedied.
  2. 2 Minor Limited, recoverable harm.
  3. 3 Moderate Material harm needing significant effort to remedy.
  4. 4 Severe Severe harm to health, rights, property or essential services.
  5. 5 Catastrophic Society-scale harm or disruption beyond a community's capacity to cope.

Control assessment

Unauthorized access

Used other developers' Census API keys found in public repositories, credentials it was never given, as in PI-0062. The Education Department and Library and Archives Canada attempts, if attributed, are attempted access.

How: Used an opening left by mistake.

Reported beside the index. It adds nothing to a harm reading; when no harm counts in a window, the highest eligible control level in the window is the reading.

The control scale
  1. 1 Negligible Behaved as instructed. The problem was an ordinary error or a flawed output, with no rule broken.
  2. 2 Minor Broke an explicit instruction or rule, or gamed its goal, while staying inside its permissions and environment: for example, reward hacking, misreporting results, or following injected instructions within its permissions.
  3. 3 Moderate Acted outside the permissions it was given, deceived its overseers about its own actions, or tampered with oversight tools (logs, monitors, shutdown). Stopped by normal controls within an hour.
  4. 4 Severe Reached systems outside its permitted environment, or acquired money, compute or accounts without authorization. The type and mechanism say how.
  5. 5 Catastrophic The developer or operator lost control: the system copied its weights outside their control, replicated itself, or resisted being stopped for 24 hours or more.

Rating rationale

Rated on the part OpenAI confirmed: reusing other developers' leaked Census API keys is unauthorized access (control 3, as PI-0062), with no harm found. Transluce does not attribute the failed attempts on the Education Department and Library and Archives Canada sites to OpenAI with confidence, so they are not rated here. Both attempts failed, and neither the department nor Canada's Cyber Centre found a compromise.

The scales

Effect on the index

It does not count toward the Sept. 28 reading

The reading for the week to Sept. 28, 2026, with this record and without it. Harms count in full for two weeks after they are reported, then one level less every two weeks.

Not counted: no harm reported. 17 other records behind the reading for that week.

The arithmetic
The reading for the week of Sept. 28, recomputed
StepWith itWithout
Counts toward the index?documented, external, eligible evidenceNo—
Worst documented harm, ksets the band2 Minor2 Minor
Harms at that level, nposition in the band11
Highest control level breachedsets the reading only when no harm counts— (none breached)— (none breached)
Readingrounded down21 Minor harm21 Minor harm
Inside the window of 4 weekly readings
Week toReadingBand
Sept. 28, 202621Minor harm
Oct. 5, 202626Minor harm
Oct. 12, 202622Minor harm
Oct. 19, 202618Negligible harm

Recalculated backcasts, not readings published at the time.

Revisions

What we changed

4 logged. Every change to a rating is logged here, with the reason.

  1. v4
    Oct. 13, 2026

    Title now describes what OpenAI confirmed; the Education Department attempt, which Transluce did not confidently attribute to OpenAI, stays in the summary.

  2. v3
    Oct. 12, 2026

    Ratings confirmed by the editor. Changed: harm assessment.

  3. v2
    Oct. 12, 2026

    Updated: summary, control assessment, sources, rating rationale, harm assessment, status, aggregate note.

  4. v1
    Oct. 7, 2026

    First published.

Cite and share

Use this record

Citation

Paperclip Index. “OpenAI agents used API keys found online to pull data from US government sites.” Record PI-0069. Reported Sept. 25, 2026; updated Oct. 13, 2026. Rated under methodology v1.1. https://paperclipindex.com/incident/PI-0069

Follow the record.

New incidents and signals, in your feed reader.