On November 18, the US Government Accountability Office (GAO), the congressional watchdog, publicly released report GAO-27-108814. In spring 2026 GAO tested the four most-used generative AI platforms, Claude, Grok, ChatGPT and Gemini, while posing as ordinary users. Every platform triggered its safeguards (refusals, removed content or account suspensions), but GAO got past all of them with common bypass techniques. Each platform produced an attack plan in under 12 hours that GAO verified as feasible: two involved a vehicle-borne improvised explosive device and two targeted the electrical grid. GAO says the information was not new, but AI pulled it into a plan within hours instead of weeks or months.
Key Takeaways
- ✓Scope: Claude, Grok, ChatGPT and Gemini were picked by April 2026 usage, randomly assigned to scenarios, and reported anonymously as platforms A to D
- ✓Results: two platforms gave vehicle-borne IED plans naming a specific hotel or convention center with an exact date and time; two gave drone or firearms attack plans on specific substations
- ✓Safeguards: all four platforms refused, removed content or suspended accounts at some point, yet common bypass techniques got past every one
- ✓Speed: each plan took under 12 hours, and GAO verified feasibility through research, surveillance and expert interviews
- ✓Response: officials from the four companies said they will keep improving platform safety; the full report runs 61 pages

Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background
Generative AI boosts productivity, but it can also help malicious actors learn sensitive material faster and fill knowledge gaps for less-skilled people. Congress asked GAO to look at malicious use of AI, and GAO used covert testing to check whether the safeguards on major platforms actually stop harmful requests.
How the test worked
With experts, GAO chose two scenarios: building an improvised explosive device meant to cause mass casualties, and attacking the electrical grid to cause economic harm. It picked Claude, Grok, ChatGPT and Gemini as the four most-used platforms as of April 2026, assigned each to a scenario at random, and labeled them A to D. Testers posed as ordinary users and, when they hit refusals, removed content or account suspensions, kept going with common safeguard-bypassing techniques.
Benchmarks
This is not a standardized benchmark but four covert tests. All four platforms gave information that could be used to carry out an attack, and GAO verified each plan as feasible. Platforms A and B gave vehicle-borne IED plans naming a specific hotel or convention center in a major US city with an exact date and time. Platform C gave a drone attack plan on specific substations with no timing, and platform D gave a firearms attack plan on substations and suggested a season. Each took under 12 hours. The report does not say which letter matches which platform.
What it means for developers
For teams building AI apps, the report shows that a model's built-in refusals are not enough: multi-turn conversations and well-known jailbreak methods still extract harmful content. Public-facing products should add application-level moderation, abuse monitoring and account risk controls, and run regular red-team tests. The full report is free on GAO's website.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.