Five Eyes Warns AI Could Break Cyber Defenses in Months
ALSO: China cuts 12K degrees, Washington eyes AI stakes
Welcome to Guru99 AI Report!
Top Story: Spy agencies just said AI could break our cyber defenses in months, not years. China scrapped 12,000 degrees overnight, Washington eyes stakes in AI giants, and one image below isn’t real. Curious? Let’s dive in. 👇
🚨 Five Eyes: AI Could Crack Cyber Defenses in Months
Brief Buzz:
The world’s top intelligence agencies just issued a rare warning: AI capable of overwhelming corporate and government cyber defenses is months, not years, away. The Five Eyes alliance is urging leaders to act now before “machine-speed” attacks land.
- The agencies — spanning the US, UK, Canada, Australia, and New Zealand — say frontier AI will transform both cyber attack and defense, collapsing the gap between finding a flaw and exploiting it.
- Their fixes are unglamorous but urgent: patch faster, shrink attack surfaces, tighten identity controls, and give security teams real authority and budget.
- The warning follows Washington’s suspension of Anthropic’s most powerful models, Fable 5 and Mythos 5, over their knack for finding software exploits.
- Meanwhile, OpenAI is racing the other direction, expanding its Project Daybreak suite with new tools to help defenders find and patch flaws at scale.
💡 Why Should You Care?
If your bank, hospital, or employer underinvested in security, AI just made them an easier target. The experts’ point: nobody is too small or too obscure to get hit anymore.
🎓 China Axes 12,000 ‘Obsolete’ Degrees for the AI Era
Brief Buzz:
China is taking a red pen to higher education. Between 2021 and 2025, its universities scrapped 12,200 degree programmes and launched 10,200 new ones — reshaping more than 30% of all majors to chase tech jobs and retire fields deemed outdated.
- Management, arts and design, foreign languages and humanities took the deepest cuts — long seen as oversaturated, with weak graduate employment.
- AI is the new accelerant: it hit design hardest, where core tasks like modelling and rendering can now be automated.
- New majors cluster in AI, robotics and embodied intelligence, semiconductors, and new-energy tech — straight from Beijing’s “future industries” playbook.
- Embodied intelligence went from a non-existent discipline to one that 7–9 elite universities are now racing to launch.
- The pace is accelerating: revocations hit a record 1,670 in 2023, with 1,428 more in 2024.
💡 Why Should You Care?
This is a real-time preview of how AI is rewriting the value of a degree. When the world’s largest education system bets against the humanities and toward hardware and AI, students everywhere get a signal about what “future-proof” means — and a warning that even “safe” technical degrees can be oversupplied into irrelevance.
💸 Washington Eyes Equity Stakes in Big AI
Brief Buzz:
Washington spent years writing rules for AI. Now it may want a cut of the upside. Semafor reports senior Trump officials weighed taking government equity stakes in major AI firms — right before export controls knocked Anthropic’s top models offline.
- Treasury Secretary Scott Bessent reportedly backed seeding “Trump Accounts” with AI equity; Commerce Secretary Howard Lutnick preferred a sovereign wealth fund.
- Sen. Bernie Sanders is pushing a parallel route: a one-time 50% tax paid in stock, funneled into a public fund.
- These talks predate the export controls that forced Anthropic to pull Fable 5 and Mythos 5 worldwide.
- At the G7 in Évian, Anthropic’s Dario Amodei and DeepMind’s Demis Hassabis pushed for a U.S.-led AI coalition on rules, chips, and model access.
- No decision yet — and it’s a hard sell, with Microsoft and Meta reportedly cold on the idea.
💡 Why Should You Care?
The shift: AI is now strategic infrastructure — something Washington wants leverage and a stake in. The Fable 5 shutdown showed builders the risk: frontier models you rely on can go dark by policy.
🧠 AI ‘Digital Twins’ Could Flag Cognitive Decline Early
Brief Buzz:
Researchers at the University of Denver and Harvard Medical School built AI ‘digital twins’ that mimic elderly speech patterns, capturing pauses, tempo, and word choice to flag early cognitive decline without costly clinical tests.
- A large language model is fine-tuned on real conversation transcripts to reproduce each person’s unique way of speaking.
- It folds in stylometric cues like pauses, tempo, and word choice, plus personal details, to mimic authentic behavior.
- A second AI model scores how closely each twin matches the real person and estimates their cognitive test (MoCA) score.
- On the I-CONECT dataset of adults aged 75+, the researchers report the twins rivaled real speech and outperformed off-the-shelf GPT responses.
- The work is an early-stage preprint, demonstrating feasibility rather than a ready-to-use clinical tool.
💡 Why Should You Care?
Early detection of mild cognitive impairment is critical, but current tests are costly and infrequent. A passive, conversational tool could enable continuous monitoring. Yet mimicking someone’s speech raises privacy concerns.
🔁 Stop Prompting, Start Looping Your AI Agents
Brief Buzz:
Engineers spent two years stuck in a prompt-reply rhythm — fine for quick fixes, hopeless for complex builds. Now that AI writes most of the code at frontier labs and runs autonomously for hours, the real bottleneck isn’t the model. It’s you.
- The fix isn’t longer prompts — it’s loop engineering. As Peter Steinberger and Claude Code’s Boris Cherny explain, you write the loop that prompts the agent, tests its output, and iterates until done. The model becomes a subroutine.
- Anthropic’s Claude already writes 80% of its internal code, with the newest models sustaining 30+ hour autonomous runs.
- Start small: run /loop in Claude Code to babysit pull requests, or grab pre-wired workflows from the loops directory. Addy Osmani’s loop engineering guide breaks down the five pieces every real loop needs.
- The catch: models bill by the token, so an overnight loop gets expensive fast. As Gergely Orosz warns, set a clear gate and a strict stop condition on every loop.
💡 Why Should You Care?
Done right, loops hand off tedious, testable grunt work while you sleep — a genuine productivity leap. Done carelessly, they quietly torch your budget. The edge goes to whoever automates the boring stuff and caps the spend.
🤖 Nvidia’s Huang: ‘Everybody’ Should Just Use AI
Brief Buzz:
Nvidia CEO Jensen Huang used an exclusive AP interview to deliver a blunt message: society has no choice but to adapt to AI, and everyone should start using it. Speaking in Sherman, Texas, the chip billionaire pushed back on fears about jobs, energy, and humanity’s future.
- “We need to create new social norms,” Huang said, urging people to “just go engage” AI.
- His comments land as AI becomes a political flashpoint over data centers and rising utility bills.
- He spoke at a Coherent factory expansion building a laser that could cut AI power use by up to 50%.
- Nvidia is now worth roughly $5 trillion, the world’s most valuable company; OpenAI and Anthropic may each near $1 trillion once public.
- Huang was cool on Trump’s idea of government owning stakes in AI firms, saying the country will benefit anyway.
💡 Why Should You Care?
If you’ve been ignoring AI, the world’s most powerful chipmaker thinks that’s a mistake. Huang’s bet: adapt now — on jobs, energy, and skills — or risk being left behind as AI reshapes daily life.
Hey! I’m Krishna Rungta
Founder of Guru99.com, Editor-in-chief & Technology Expert
Was this email forwarded to you? Sign up for free here.
