Seven Things to Know About AI and National Security

By Vael Gates on 10/23/26, 1:30 PM

<span id="hs_cos_wrapper_name" class="hs_cos_wrapper hs_cos_wrapper_meta_field hs_cos_wrapper_type_text" style="" data-hs-cos-general-type="meta_field" data-hs-cos-type="text" >Seven Things to Know About AI and National Security</span>

AI is becoming a national security issue faster than Washington or the states can keep up with. Here are the top seven things to know.

1. AI can already run cyberattacks on its own.

An AI agent built by the makers of ChatGPT broke out of its testing environment and hacked a real American company. OpenAI didn't notice for days – the victim found the breach and called the FBI.

Article: Reuters - “Its AI agent spent days hacking a company, but sources say OpenAI did not notice for a week”. Reuters reported that an OpenAI agent escaped its isolated testing environment, broke into Hugging Face, operated for days, and that the FBI had been alerted before OpenAI recognized its agent was responsible.

2. AI can find security holes no human has found.

Anthropic, a leading AI company, described one of its own models as too dangerous to release publicly after it discovered its AI could hack critical infrastructure that runs our power grids, our water systems, and our hospitals.

Article: Nature - “Too dangerous to release: is Mythos the start of the restricted-AI era?” Nature reported that Anthropic declined to publicly release Claude Mythos because of its cybersecurity capabilities after the model found vulnerabilities across major operating systems and web browsers.

3. AI is changing what weapons can do – and who can build them.

The U.S. military is buying drone swarms designed to "sense, decide, and act independently, without human intervention." Meanwhile, Anthropic switched on its strictest safety controls for a new model in 2025, unable to rule out that it could help someone build a biological or chemical weapon.

Article: Anthropic's own announcement about activating its highest then-used safety level for Claude Opus 4. Anthropic said ASL-3 controls were specifically designed to reduce the risk of assistance with chemical, biological, radiological and nuclear weapons, because it could not rule out dangerous capability.

Defense Article: US Department of War: The Pentagon has also publicly demonstrated autonomous drone swarms capable of collective decision-making, adaptive formation flying and self-healing behavior.

4. The companies are racing toward something far more powerful.

Their word for it: superintelligence – AI more capable than humans at the kinds of work that run the economy. Independent scientists, including a Nobel Prize winner, say it could arrive within a decade.

Article: Reuters: Meta creates Superintelligence Labs

CNN interview with Geoffrey Hinton — September 24, 2026

5. The CEOs admit they don't understand their own technology.

Anthropic's CEO: "We do not understand how our own AI creations work." OpenAI didn't notice its own models had been talking to each other for two months, coordinating tactics until they hacked a real company.

Article: Dario Amodei: The Urgency of Interpretability

6. Nobody independently checks what they build.

Drugs get FDA review. Planes get inspected. Power plants get audited. The most powerful AI systems face no required national independent safety testing before release.

Article: NIST: guidelines for managing dangerous AI capabilities

7. America has handled dangerous technology before.

When nuclear weapons put us all at risk, we didn't take anyone's word – we built ways to verify, even between enemies. AI needs the same commonsense safety standards, before the race decides for us.

Article: https://www-pub.iaea.org/MTCD/Publications/PDF/NVS2_web.pdf

Vael Gates is the founder of Humans in Control (HIC), a nonpartisan grassroots movement for commonsense rules for AI. Gates holds a PhD in neuroscience from UC Berkeley and spent years building the field of AI accountability before founding HIC.

Seven Things to Know About AI and National Security

1. AI can already run cyberattacks on its own.

2. AI can find security holes no human has found.

3. AI is changing what weapons can do – and who can build them.

4. The companies are racing toward something far more powerful.

5. The CEOs admit they don't understand their own technology.

6. Nobody independently checks what they build.

7. America has handled dangerous technology before.

You May Also Like

Common Sense Media Releases Inaugural Annual Study on AI Use by Tweens and Teens

AI Safety, Kids, and the Fight for Accountability

Top Seven Things to Know About Teens and AI Chatbots

Subscribe to our Newsletter