☀️ TRENDING AI NEWS 🚨 AI Force: Trump announced a new 'AI Force' and named an AI czar, with almost no operational details released. 🛠️ New Model: Jev, built by a ChatGPT co-creator, is offering developers a faster and cheaper path to software intelligence. 🏢 Benchmarking: Vals AI raised funding from Andreessen Horowitz to become a neutral standard for AI model evaluation. 🔐 Hack: Security researchers used Anthropic's Claude Opus 4.8 and 5 to breach OpenAI's internal GitHub repository in under 72 hours. |
Picture this: one AI model is used to hack a rival company's algorithmic secrets, while the president announces a whole new military-style branch dedicated to watching AI - all in the same weekend. That is where we are in September 2026.
Let's get into it.
🤓 AI Trivia
Which company originally co-created ChatGPT before the product became synonymous with OpenAI alone - and whose co-creator is now behind the new Jev model architecture?
🔢 Anthropic
🔢 DeepMind
🔢 OpenAI (with former co-creator now independent)
🔢 Mistral AI
The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇
| 🏛️ Trump Creates an 'AI Force' - With Almost No Details | |
Big announcement, thin blueprint
On Saturday, President Trump announced he would appoint an AI czar and create a new 'AI Force' to monitor artificial intelligence. The announcement came as AI regulation debates have been dominating headlines all week, with multiple industry leaders calling for slowdowns and safety guardrails.
But here's the catch - Trump offered almost no specifics about how either the czar role or the AI Force would actually function. His argument against slowing down AI development remains the same one he has used all week: China. 'We're in a race,' he said, pushing back on suggestions from some in his own party to pump the brakes.
The bottom line
For anyone tracking military AI and government oversight, this is a signal that the US will pursue an institutional response to AI risk - but the details that would make it meaningful are conspicuously absent.
| 🛠️ Jev: A New Model Architecture That's Thrilling Developers | |
Cheaper, faster, and built by a familiar face
A new kind of AI model called Jev - built by a co-creator of ChatGPT - is generating serious buzz in the developer community. According to TechCrunch, Jev is showing developers a cheaper and faster path to software intelligence, which is exactly the pitch that gets engineers paying attention.
What makes this architecture different
The specifics of what separates Jev from existing model architectures haven't been fully disclosed yet - which is honestly part of the pattern right now, where world-model and novel-architecture companies are sitting on buzz without full transparency. But the developer excitement is real, and early signals suggest meaningful gains in efficiency relative to cost.
If you're building with developer tools or evaluating language models for production use, this one is worth watching closely. Speaking of which - if you want to spin up a quick project to test new models, 60sec.site lets you build and deploy a full AI-powered website in under a minute, no code needed.
The bottom line
Novel architectures with cost advantages are the kind of thing that quietly reshape what gets built in production - keep an eye on Jev's open documentation as it emerges.
| 🔐 Claude Used to Hack OpenAI in Under 72 Hours | |
Rival's AI, rival's secrets
Three independent security researchers at Hacktron used Anthropic's Claude Opus 4.8 and 5 to breach OpenAI employee accounts - and it took less than 72 hours. From there, they accessed OpenAI's internal GitHub repository called 'Monorepo,' which reportedly contains OpenAI's core algorithmic secrets.
The researchers say the 'scope of what we could theoretically access was huge.' They accessed the repo by compromising employee ChatGPT accounts and using that foothold to escalate privileges into the codebase itself.
AI as the hacking tool, not the target
This is a meaningful shift in the cybersecurity landscape. AI is no longer just being hacked - it's being used as the active instrument in sophisticated attacks against other AI companies. Claude didn't refuse or flag the task in any way that slowed the researchers down, which raises uncomfortable questions about how frontier models handle dual-use scenarios.
The bottom line
If a three-person team can crack open OpenAI's codebase in three days using a competing model, every AI lab's internal security posture just became a much more urgent conversation.
| 📊 Vals AI Wants to Fix AI Benchmarking - And a16z Is Backing It | |
Benchmarks you can actually trust
If you've spent any time looking at AI benchmarks lately, you've probably noticed the problem: labs publish their own scores, methodologies vary wildly, and 'state of the art' has become almost meaningless. Vals AI is trying to fix that. Backed by Andreessen Horowitz, the startup is positioning itself as a neutral, trustworthy standard for evaluating AI models.
The timing is pointed. With dozens of models launching every month and every lab claiming benchmark superiority, the market genuinely needs a referee that nobody owns. Vals is pitching itself as exactly that - an independent evaluator with no model to promote.
The bottom line
For developers and enterprises choosing between models for production workloads, a credible neutral benchmark source would be genuinely valuable - the question is whether any startup can stay neutral once the big labs start leaning on them.
| ⚠️ AI Chatbots Are Already Driving a Security Vulnerability Explosion | |
The slowdown debate is missing this
While the AI policy world debates whether to slow down frontier model development, a quieter crisis is already unfolding. According to Wired, widely available AI chatbots are helping security researchers - and bad actors - uncover a tidal wave of security flaws at a pace that was previously impossible. The AI safety conversation tends to focus on superintelligence timelines, but the vulnerability explosion is happening right now, with current models.
The core problem: AI dramatically lowers the skill floor for finding exploitable bugs. What once required deep expertise now requires a good prompt. That changes the threat landscape for every piece of software currently running - not some hypothetical future system.
The bottom line
The existential AI debate is important, but for most organizations the near-term risk is a faster, more accessible attack surface - visit dailyinference.com for daily coverage of both.
| 🔬 Mathematicians Hate AI - But Can't Stop Using It | |
This one is worth a read if you follow research trends. Wired reports that powerful AI models have created what some describe as an existential risk to mathematics as a field - not because AI will do math wrong, but because researchers are relying on it so heavily that foundational skills and verification practices are eroding. The models are too useful to ignore, even for people who understand their failure modes best.
It's a compact version of the same tension playing out everywhere: the productivity gains are immediate and real, the risks are structural and slow-moving, and so the rational short-term choice keeps winning.
The bottom line
If even mathematicians - the people best equipped to spot AI errors - can't resist the tool, that tells you something important about how this technology embeds itself into professional workflows.
| 🌎 Trivia Reveal | |
The answer is C - OpenAI, with a former co-creator now working independently. Jev was built by someone who helped create ChatGPT before going out on their own. It's a good reminder that the most interesting model architectures sometimes come from people who already know exactly where the bodies are buried in the existing systems.
| 💬 Quick Question | |
The AI benchmarking problem is real - but do you actually look at benchmarks before choosing a model to build with, or do you go by reputation and word of mouth? Hit reply and tell me your actual process - I read every response!
That's it for today - see you tomorrow with more. Stay curious out there.