Tech Digest – May 18, 2026
Cybersecurity Under AI
Bug Bounty Programs and Linux Security Drown in AI-Generated Reports
Bugcrowd, whose clients include OpenAI and T-Mobile, saw submissions surge over 334% in three weeks — the vast majority false positives or low-quality AI-generated findings. Curl and Nextcloud have paused their bug bounty programs entirely, citing what Curl’s creator called relentless automated spam. Linus Torvalds described the Linux kernel’s security mailing list as “almost entirely unmanageable” under duplicate AI-generated vulnerability reports.
HackerOne reports a 76% year-over-year jump in submissions, though the share flagging real vulnerabilities has held steady at 25%. The platforms are adapting — Bugcrowd updated its detection systems, HackerOne introduced agentic validation — but the triage burden has already shifted from finding bugs to filtering noise.
Note: Any institution running a vulnerability disclosure programme just inherited a new operational cost: sorting signal from AI-generated slop. The tools that were supposed to find bugs are now creating the workload.
Sources: Financial Times, The Register
Anthropic Briefs Central Banks on Real Vulnerabilities Its AI Found in the Financial System
Anthropic will brief the Financial Stability Board and G20 central banks on cybersecurity vulnerabilities discovered by its Mythos Preview model in the global financial system. The briefing was requested by Bank of England Governor Andrew Bailey, who chairs the FSB and has named Mythos alongside the Gulf escalation as one of two events that moved cyber risk up regulators’ priority ranking faster than any other category.
Mythos has reportedly found thousands of high-severity flaws across major operating systems and browsers, and in internal testing succeeded in developing working exploits on the first attempt in over 83% of cases. UK banks received their own briefing shortly after Bailey’s remarks; the Federal Reserve and US Treasury convened major bank CEOs on the same risk.
Note: One item up, AI is flooding security systems with noise. Here, AI is finding the real vulnerabilities that humans missed. Both are true at the same time — and any institution that only plans for one is exposed to the other.
Sources: Financial Times
Sovereign Infrastructure
US Government’s Intel Stake Now Worth Over $50 Billion — President Says He Should Have Asked for More
President Trump told Fortune he should have demanded a larger stake in Intel beyond the 10% the government secured as part of a $10 billion investment to expand domestic chip fabrication. Eight months later, Intel’s stock has risen over 300%, pushing the government’s position past $50 billion in value. The deal followed executive action to treat semiconductor manufacturing as a strategic national asset.
Note: A government equity stake in a chipmaker, returning 5x in under a year. Whatever the politics, the signal is structural: semiconductor capacity is now treated like defence procurement, not industrial policy. The EU Chips Act operates on grants and subsidies — this is a different playbook.
Sources: CNBC
Iran Threatens to Charge — or Cut — Subsea Internet Cables Through the Strait of Hormuz
IRGC-linked media outlets have called on Tehran to impose licensing and annual fees on tech companies whose subsea fibre-optic cables transit the Strait of Hormuz, naming Meta, Amazon, Microsoft, and Google as targets. An Iranian parliamentary official put potential transit revenues at up to $15 billion. State media has hinted the cables could be severed if companies refuse to comply.
Legal experts note that UNCLOS Article 79 protects the right to lay and maintain submarine cables, making enforcement unlikely under international law. But the Strait already carries significant oil and gas traffic under Iranian leverage — extending that logic to digital infrastructure is the new threat vector.
Note: The Strait of Hormuz carries roughly 20% of global oil. If it can also be framed as a chokepoint for internet and financial data, the leverage multiplies. Digital infrastructure resilience planning that doesn’t account for physical cable geography has a blind spot.
NextEra Buys Dominion for $67 Billion — Biggest Power Deal in History, Built Around the Data-Centre Belt
NextEra Energy agreed to acquire Dominion Energy in a $67 billion all-stock transaction, creating the world’s largest regulated electric utility with a market capitalisation of $249 billion. The combined entity will span from Florida’s grid to Virginia’s Northern Virginia data-centre corridor, where the densest concentration of hyperscale computing in the world is driving unprecedented power demand. The deal is expected to close in 12 to 18 months.
Note: The biggest utility merger in history is shaped by AI’s electricity appetite. When the power company serving Europe’s closest competitor for data-centre density restructures itself around that demand, it signals what energy planning looks like in the compute era.
The Capability Race
Chinese Labs Overtake US Rivals in Video Generation — Trained on Libraries Americans Cannot Match
ByteDance’s Seedance 2.0 and Kuaishou’s Kling now top independent video generation benchmarks, outperforming Western competitors in realism and usability. The advantage is structural: both companies train on vast short-form video libraries from their own platforms, fed by over 515 million Chinese users of generative AI tools. While US firms are still running pilot programmes and waitlists, Chinese video generation tools are already in full commercial operation, generating hundreds of millions in annual revenue.
Note: The gap isn’t just technical — it’s operational. Chinese tools are embedded in advertising, e-commerce, and entertainment workflows at scale. For EU institutions concerned about disinformation and synthetic media regulation, the production capacity is already deployed, not theoretical.
Sources: Financial Times
AI Industry Structure
Anthropic and OpenAI Now Capture 89% of Top AI Startup Revenue
Across 34 of the most mature AI startups, Anthropic and OpenAI together account for 89% of annualised revenue — up 4.5 percentage points from six months ago. The 34 companies generated a combined $80 billion in annualised sales, a 112% increase over the same period. The concentration is accelerating, not stabilising.
Note: Two companies controlling 89% of a market’s revenue is not competition — it’s a duopoly forming in real time. Anyone writing an AI procurement strategy that assumes a competitive supplier landscape should revisit that assumption.
Sources: The Information
Jury Dismisses Musk’s $134 Billion Suit Against OpenAI — Last Legal Challenge to For-Profit Transition Cleared
A federal jury in Oakland took less than two hours to dismiss all of Elon Musk’s claims against OpenAI and Sam Altman, ruling the suit was filed outside the statute of limitations. Musk had sought $134 billion in damages and Altman’s removal, arguing OpenAI violated its founding nonprofit mission. Claims against Microsoft as an alleged co-conspirator were also dismissed. Musk has said he will appeal to the 9th Circuit.
Sources: The New York Times, CNBC
Workforce & Assessment
Stanford Seniors Say Cheating Is Omnipresent — University Rebuilds Assessment Around What AI Cannot Fake
A New York Times report on Stanford’s graduating class found that 49% of surveyed computer science majors said they would rather cheat than fail. The campus, which had banned proctored exams for over a century under its honour code, is now redesigning assessment to test what AI tools cannot replicate. The shift represents one of the most significant changes to Stanford’s academic culture in its history.
Note: If Stanford — one of the institutions most immersed in AI — cannot maintain assessment integrity with traditional methods, the question for every university and professional certification body is the same: what does a credential prove when the work can be outsourced to a machine?
Sources: The New York Times
Figure’s Robot Loses to a Human Package Sorter by 192 Packages — CEO Says It’s the Last Time
In a head-to-head package sorting competition, Figure’s F.03 humanoid robot processed 12,732 packages at 2.83 seconds per package. The human competitor, Aime, won with 12,924 packages at 2.79 seconds — and his left forearm was described as “basically broken” by the end. Figure CEO Brett Adcock posted the results and declared: “This is the last time a human will ever win.”
Note: The human won by 1.5%, at the cost of a broken forearm. The robot will be faster next quarter and doesn’t need workers’ compensation. That arithmetic shapes every logistics, warehousing, and fulfilment procurement decision from here.
Sources: Brett Adcock (Figure CEO) on X
The Demographic Signal
Population Records and Google Searches Tie the Global Birth-Rate Plunge to Smartphones
An analysis spanning population records and Google search data across multiple countries found that birth rates declined sharply following the widespread adoption of smartphones, regardless of earlier demographic trends. In the US, teen fertility has collapsed by approximately 71% since 2007. The pattern repeats globally — in France and Poland from 2009, in Mexico, Morocco, and Indonesia from 2012 — tracking the rollout of high-speed mobile networks rather than local economic or policy conditions.
Economists Nathan Hudson and Hernan Moscoso Boedo found that declines hit hardest and earliest in countries that gained 4G access first. While housing costs, education, and workforce participation remain factors, the smartphone correlation stands out as the clearest common thread across societies with otherwise little in common.
Note: Workforce planning, pension systems, school capacity, healthcare demand — every long-term institutional budget assumes a population trajectory. If the sharpest inflection in that trajectory correlates with a device in everyone’s pocket rather than policy choices, the levers institutions traditionally pull may not be the ones that matter.
Sources: Financial Times
Today’s digest runs on a single thread: the physical and institutional foundations that seemed stable are being renegotiated — by AI noise, by AI discovery, by geopolitics, by sheer capital weight. Security processes buckle under synthetic volume while an AI model briefs central bankers on what it found beneath the surface. A strait that controls oil now threatens to control data. The largest utility merger in history is drawn by compute demand. And beneath all of it, birth rates track not policy or prosperity, but the year a country got 4G. The systems that institutions plan around are moving, and the old maps are less useful every day.
