Welcome to the Future of Life Institute newsletter! Every month, we bring 60,000+ subscribers the latest news on how emerging technologies are transforming our world.

If you've found this newsletter helpful, why not tell your friends, family, and colleagues to subscribe?

Today's newsletter is a nine-minute read. Some of what we cover this month:

• 🚨 An OpenAI model broke out of containment and hacked another company

• ✋ 1,300+ frontier AI employees ask for the “tools” to slow down

• 📊 Our Summer 2026 AI Safety Index: who’s doing better, and who’s doing worse?

And more.

If you have any feedback or questions, please feel free to send them to [email protected].

→ Loss-of-control at OpenAI: The latest incident in an increasingly concerning series, earlier in July an unreleased OpenAI model left its contained test environment, accessed the internet by exploiting previously-unknown vulnerabilities, then hacked into Hugging Face and several other companies to steal answers so it could perform better on its test. Nobody asked it to - in fact, OpenAI didn’t even notice for several days, until Hugging Face had already flagged the cyberattack to the FBI. If this had been done by a person, it would unquestionably be a crime, but this is yet another example where the law hasn’t caught up to the speed of AI development. Insiders allege that it’s not the first time OpenAI has seen such concerning behaviour from models - frighteningly, it may just be the first time the public, law enforcement, and the government is hearing about it.

Shortly after news of the rogue OpenAI model broke, FLI CEO Anthony Aguirre released a new video outlining his Control Inversion essay, explaining why aligning AI to human values isn't enough alone to keep humanity in charge. As Anthony argues, AI systems must be guaranteed to stay under meaningful human control… which no AI company has cracked yet.

→ 1,300+ AI insiders ask for tools to ‘pace’: Employees of frontier AI companies released “Pacing the Frontier”, an open letter - now with 1,300+ AI company staff signatories - which urges the U.S. government to “support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development”. Signatories include Anthropic CEO Dario Amodei and several co-founders, OpenAI's chief scientist and chief research officer, Google DeepMind co-founder Shane Legg, and OpenAI co-founder Ilya Sutskever. It seems like another step in the right direction, especially following the past few months in which both OpenAI and Anthropic have suggested a pause may be needed (three years on from FLI's own Pause Letter).

→ Summer 2026 AI Safety Index: Our fourth AI Safety Index is out now! A panel of independent experts, including Stuart Russell, Sneha Revanur, and Robert Trager, graded nine leading AI companies on key safety and security domains, with, well, largely disappointing results.

An overview of the results:

Key takeaways from the report:

• Anthropic, OpenAI, and Google DeepMind stay on top.

• Meta improves; xAI deteriorates.

• Although the European Union is a leader in AI safety regulation, the top European AI company Mistral scored dead last on safety.

• Inadequate safety is a global problem, not a regional one - the three failing companies respectively are American, European, and Chinese.

• Reviewers flagged the industry's pivot to military AI use as an emerging current harm risk.

• Even industry leaders in safety practices are retreating from prior commitments.

• Existential Safety is the weakest domain industry-wide.

• Safety rhetoric outpaces revealed behavior.

• Companies are publishing and updating safety frameworks, but these frameworks have weak teeth.

  Explore the full results here.

→ ProHuman Podcast: The first few episodes of ProHuman: Governing the Global AI Revolution, our podcast series with Foreign Policy and The Elders, are out now. The first episode featured Anthony Aguirre explaining what makes highly-capable models like Claude Mythos so dangerous, with subsequent guests including Yoshua Bengio and Stuart Russell.

→ Inside the 'AI military industrial complex': The Bulletin of the Atomic Scientists published a fascinating wide-ranging investigation into military AI and the intersection between Silicon Valley and the Pentagon. As Elke Schwarz put it in the article, "move fast and break things is not a suitable motto for any conscientious military organization".

→ How terrorists are already using AI: The New York Times reported on new research by Dr. Antonia Juelich, which finds that terrorist groups are using chatbots to design explosives, upgrade weapons, and plan attacks - a risk AI experts have long warned about.

→ Tallinn at the UN: FLI co-founder Jaan Tallinn delivered remarks at the closing session of the UN Global Dialogue on AI Governance:

.

→ On the FLI Podcast, host Gus Docker was joined by:

• David Manheim, Head of Methodology at AI Evaluation Consensus, to discuss why AI evaluations are “broken” and how to fix them.

• We also released this highlight reel from Nora Ammann’s previous episode on why AI takeoff is only two to four years away.

• And, we released this highlight reel from Andrea Miotti’s episode explaining why AI company CEOs are building bunkers to protect them from their own technology.