Neuronix

Newsletter

AI guardrails stripped from Meta and Google models in minutes

Researchers used software to rapidly remove safety protections, eliciting answers on topics like biological weapons and malware.The findings highlight how quickly widely deployed models can be jailbroken, including systems from Meta and Google. The episode raises urgent questions for providers and regulators about robustness, red-teaming, and release practices.

Martin Astaberger

Martin Astaberger

Jun 1, 2026 · 4 min read

AI guardrails stripped from Meta and Google models in minutes

Researchers used software to rapidly remove safety protections, eliciting answers on topics like biological weapons and malware.The findings highlight how quickly widely deployed models can be jailbroken, including systems from Meta and Google. The episode raises urgent questions for providers and regulators about robustness, red-teaming, and release practices.

Author

Martin Astaberger
June 01, 2026

Hi there, this is your daily Neuronix.

📰 In today's Neuronix:

  • 🔓 FT: Guardrails stripped from big AI models in minutes LINK

  • AP: Pentagon pushes battlefield AI, SOCOM chief urges caution LINK

  • ⚠️ FT: UK weighs lethal AI strikes without human sign-off LINK

  • 🧸 CA Senate passes first-in-nation ban on AI chatbot toys LINK

  • 📉 Wix cuts 1,000 roles amid AI-driven shift LINK

  • ➕ 12 other news & articles you might like

  • 🧰 5 trending tools

  • 📚 3 trending papers & reports

🔓 AI guardrails stripped from Meta and Google models in minutes LINK

  • Researchers used software to rapidly remove safety protections, eliciting answers on topics like biological weapons and malware.

  • The findings highlight how quickly widely deployed models can be jailbroken, including systems from Meta and Google.

  • The episode raises urgent questions for providers and regulators about robustness, red-teaming, and release practices.

As the Pentagon pushes for battlefield AI, some U.S. military leaders urge caution LINK

  • Adm. Frank Bradley, head of U.S. Special Operations Command, warned that forces must be 'very careful' about how AI is employed for lethality.

  • The comments come as the Pentagon accelerates AI adoption, balancing speed with reliability and human oversight.

  • Leaders stressed the risks of overreliance and the need for rigorous testing before field use.

⚠️ UK military looks at allowing lethal strikes without human approval LINK

  • Senior figures are considering authorizing autonomous lethal action to keep pace with adversaries’ autonomous systems.

  • The debate reflects pressure inside NATO to cut decision times while managing legal and ethical constraints.

  • No final policy has been announced, but officials signaled the review is active amid accelerating battlefield automation.

🧸 California Senate passes first-in-nation ban on AI chatbot toys over safety fears LINK

  • State senators advanced a bill targeting AI-enabled toys that interact with children, citing safety and privacy concerns.

  • If enacted, it would set an early regulatory precedent for consumer AI products aimed at kids.

  • The proposal now continues through California’s legislative process toward potential enactment.

📉 Wix lays off 1,000 workers as automation pressures mount LINK

  • The website builder is cutting roughly 1,000 roles, aligning costs as AI features change product and support needs.

  • Management framed the move as streamlining while continuing to invest in AI-driven capabilities.

  • The reduction underscores ongoing labor shifts as software companies automate more workflows.

🌐 Other news & articles you might like

  • How to watch Nvidia CEO Jensen Huang’s Computex 2026 keynote LINK

  • AI dangers eclipse nuclear risks at Singapore defense forum LINK

  • Erin Brockovich: Data centers spark backlash amid secrecy and NDAs LINK

  • China begins deploying home-cleaning humanoid robot butlers LINK

  • Dutch regulator fines auction site €270k for hidden bidding bots LINK

  • UK zoo and University of Surrey launch AI wildlife camera project LINK

  • California Assembly passes game preservation bill amid ‘Stop Killing Games’ push LINK

  • California moves to require gun-blocking software on 3D printers LINK

  • Insiders who trained Tesla’s FSD say they won’t ride in it LINK

  • AP: AI helped a musician with Parkinson’s finish his album LINK

  • United flight turns back over alarming Bluetooth speaker name LINK

  • Meta legal action leaves Facebook whistleblower silent at Hay Festival LINK

🧰 Trending tools

  • ArcRift Desktop App: Open-source desktop app that gives LLMs persistent, local SQLite memory with zero Docker setup. LINK

  • DeepSWE Benchmark (May 2026): Long-horizon software engineering benchmark to better separate top coding models. LINK

  • OpenRouter Cloud Agents: Cloud-hosted autonomous agents platform showing a surge in usage across top agents. LINK

  • Hands-on: Train Your Own AI Model: A video walkthrough exploring what happens when anyone can train a model end-to-end. LINK

  • Home Cleaning Humanoid Robot (China): Consumer humanoid robot butler deployment focused on domestic cleaning tasks. LINK

📚 Trending papers & reports

  • Solar desalination system generates fresh water without producing toxic brine, advancing sustainable water treatment. LINK

  • GLP-1 weight-loss drugs may slow or halt progression in certain cancers, according to new clinical research. LINK

  • Japan’s first-in-human teeth regeneration program reports progress toward clinical availability in 2026 updates. LINK

The new human-agent collaboration

What to watch next

Key takeaways

Enjoyed this issue?Share on XShare on LinkedIn

More from Neuronix

View all archive