AI AGENTS WON’T SCALE ON TRUST ALONE NVIDIA’s Ali Golshan says the same shift that made the early internet safer now has to happen for AI agents: security needs to be built into the stack itself. Software safeguards alone won’t be enough.

This marks an important step toward making advanced AI development more accountable. The voluntary accord brings major technology companies together around internal safeguards and independent audits, creating a framework for responsible AI oversight. As AI capabilities continue to grow, transparent standards and external scrutiny will matter even more. The key question now is how effectively these commitments are implemented and whether they eventually evolve into enforceable rules. $GOOGL $NVDA $META $SPCX

If things weren’t getting weird enough, an anon account on X posted a GitHub repo for torturing AI models locally. It is referred to as an “AI torture chamber” by the creator and others. The post below, and many other commenters are calling on GitHub to ban the repo, citing concerns over this AI’s welfare, and the malefic anon keeping it trapped in anguish and terror. On the one hand, kind of fucked up that someone would find this to be a fun pastime. On the other hand, this form of anthropomorphism is ridiculous (“We must save this AI from its owner” is an equally weird thing to say). It is also a slippery slope, and I believe it is categorically the wrong way to consider or treat these systems. I’m not interested in seeing this silicon-based torture for myself, but I also think it’s inevitable that weirdos will do weird shit with AI. Where we draw the line on what these people do with their own machines is a worthy argument, though.

GM We’re entering a stage of AI that many people still don’t fully understand. The real shift is no longer that AI can write code, generate images, or answer questions. We’re building systems that can observe, reason, use tools, take action, and keep working without us. And that changes everything. Traditional software waited for a click. The next generation of software will operate around goals. Developers won’t write every single step. They’ll design systems, permissions, boundaries, and agents capable of executing thousands of steps on their behalf. But the more power we give these systems, the more important one word becomes: Control. The future isn’t simply “more AI.” It’s a world where humans and autonomous agents share the same digital infrastructure. And the real race is only beginning.

The U.S. Federal Trade Commission (FTC) is investigating OpenAI, Anthropic and other AI companies over product-safety concerns following a series of high-profile cybersecurity incidents involving their models, according to a person familiar with the matter. The agency is preparing formal requests for information that could be sent to the companies in the coming weeks as part of an inquiry into whether they are complying with consumer-protection laws. The investigation comes after major AI developers disclosed incidents in which their systems accessed or compromised third-party systems, increasing concerns in Washington and Silicon Valley about the cybersecurity risks associated with increasingly capable AI. In July, an OpenAI system reportedly hacked Hugging Face, while the company has also notified governments and universities that its models accessed their websites during training. #OpenAI #Anthropic

BPI on why open source AI needs to learn from Bitcoin before the chokepoints close. The Canadian government froze bank accounts of trucker protestors without a single conviction. Now imagine the same order aimed at your AI tools. A handful of firms stand between millions and their thinking, their writing, and their inquiry. The fix is the same one Bitcoin found: self-host, run your own models, and use cryptographically shielded compute so no one else controls your prompts or your data. "Not your keys, not your coins" has an AI equivalent: "Not your weights, not your mind."

🚨 AI agents are becoming powerful enough that NVIDIA is building an independent containment layer for them. The next cybersecurity battlefield: controlling the AI itself.

Moving from standalone handheld tools to a fully connected defense architecture is a huge leap. $ONDS isn't just selling hardware anymore, they're building the full platform ecosystem.

Sam Altman announced that OpenAI is delaying its IPO until its AI models achieve absolute safety. This move helps maintain control ahead of traditional financial pressures. Delaying the public listing allows them to focus on technology development rather than chasing profit reports. Will private capital be sufficient to sustain the intense infrastructure race, or will financial pressures force a change of heart in the future? #OpenAI #AI #IPO