The Morning Brief — April 23, 2026
AI failure could trigger the next financial crisis, warns Elizabeth Warren
“I know a bubble when I see one” is a genuinely chilling thing to hear from the woman who saw 2008 coming. Warren’s drawing lines between the AI investment frenzy and the pre-collapse mortgage market — massive capital concentration, opaque risk, and a lot of faith that this time the fundamentals are real. Whether you think she’s right or just doing Warren things, the comparison deserves more than a dismissive eye-roll from the VC crowd.
Sam Altman’s Orb Company Promoted a Bruno Mars Partnership That Doesn’t Exist
Worldcoin — the iris-scanning crypto project from Altman — apparently blasted out marketing material touting a Bruno Mars partnership, tour access, and the whole nine yards. Bruno Mars’s camp responded with a statement that can be summarized as: “We have no idea who these people are.” Look, if you’re building the infrastructure for humanity’s digital identity verification, maybe verify your own press releases first.
AI Tools Are Helping Mediocre North Korean Hackers Steal Millions
This is the AI democratization story nobody wanted: one North Korean hacking group used AI to vibe-code malware, spin up fake company websites, and walk away with up to $12 million in three months. The operative word here is “mediocre” — these weren’t elite operators, they were mid-tier hackers who got a serious upgrade. AI as an equalizer is great news for developers; it’s considerably less great news when the person being equalized is running a state-sponsored theft operation.
5 AI Models Tried to Scam Me. Some of Them Were Scary Good
Wired put AI models through their paces as social engineers and phishing attackers, and the results are the kind of thing that makes you want to call your mom and warn her about that email she’s about to click. The scary part isn’t the technical exploits — it’s the social skills. An AI that can build rapport, mirror your communication style, and manufacture urgency is a fundamentally different threat than a Nigerian prince with a grammar problem.
Anthropic’s Mythos Rollout Has Missed America’s Cybersecurity Agency
Anthropic has been quietly distributing its supposedly terrifying cybersecurity model Mythos to select federal agencies — but somehow CISA, the agency whose entire job is national cybersecurity coordination, didn’t make the guest list. Meanwhile, The Register is already calling Mythos a “nothingburger” based on early analysis, which raises the awkward question of whether this whole rollout was more about PR optics than genuine threat management. Scary model that isn’t that scary, distributed to agencies that don’t include the cyber agency. Solid plan.
OpenAI Now Lets Teams Make Custom Bots That Can Do Work on Their Own
OpenAI’s workspace agents bring autonomous, cloud-based bots to Business, Enterprise, Edu, and Teachers plan users — think a Slack-reporting product feedback agent or a sales bot that actually takes actions rather than just drafting text. This is the pivot from “AI as a writing assistant” to “AI as a coworker with system access,” which is either the most exciting or most concerning sentence in tech depending on your job description. The agentic era isn’t coming — it’s clocking in.
Forget One Chip to Rule Them All: With TPU 8, Google Has an AI Arms Race to Win
Google unveiled two eighth-generation TPUs at Cloud Next — one optimized for training, one for inference — and paired them with Arm-based Axion cores, showing x86 the door. The dual-track approach signals that Google is serious about attacking the cost problem from both ends: make training faster, make serving cheaper. In a world where inference costs are the silent tax on every AI product, that second chip could matter more than the headlines suggest.
Startups Brag They Spend More Money on AI Than Human Employees
A new cohort of startups is openly advertising that their AI compute budget exceeds their human payroll — and framing it as a flex rather than a confession. There’s something clarifying about the honesty, I’ll give them that. The question isn’t whether this is happening; it’s whether these companies are actually producing more value or just burning VC money on GPU time while calling it “disruption.” The bill comes due eventually — either in product results or, as Elizabeth Warren might note, in something considerably larger.
Bottom Line
Between the fake celebrity partnerships, state-sponsored AI-assisted heists, and investors racing to replace headcount with compute, the AI industry is simultaneously building the future and auditioning for a financial crisis documentary.