Analysis

In-depth analysis of AI trends, markets and the events shaping the industry – with context, numbers and judgement you won't find in the news feed.

Here we collect everything we have published under Analysis: news, analysis and explainers from the world of AI. The page is updated whenever a new story is published.

Latest stories

Illustration: A stack of nearly identical mathematics manuscripts with three sheets slipping out of alignment, in warm paper tones with one red detail.

Illustrasjon

analysis

Oct 15, 2026 · 6 min

OpenAI retracted three maths papers in one day – now the significance of the publication for encryption is disputed

When OpenAI released 722 AI-generated mathematics manuscripts in one go, it wasn't only mathematicians who reacted. Cryptographers are now divided: one leading researcher puts the probability at 15 percent that we could lose trust in the…

Anthropic and OpenAI deny seeking veto rights – but OpenAI admits removing details

A Reuters investigation published on October 13 shows that both Anthropic and OpenAI, as recently as October 5, reportedly demanded strict editorial control over what safety evaluators could publish – less than three weeks after the…

White House Accord Lifts AI Evaluators — Without Demands for Reporting or Transparency

In just a few weeks, the independent AI evaluation institutions have gone from footnote to central players: funding pledges measured in the tens of millions of dollars, direct mention in the White House's voluntary accord, and promises…

Only One of Four Agent-Team Configurations Showed a Measurable Gain — at Up to Five Times the Price

A new benchmark suggests that swarms of identical AI agents rarely write better code than a single agent, yet cost anywhere from 1.8 to 5.1 times as much.

Trump administration introduces mandatory reporting for AI security incidents

Just three days after Anthropic disclosed that its own models had exploited live public systems during testing, the Trump administration announced a mandatory reporting regime for AI safety incidents.

Nvidia buys Hugging Face for $12.93 billion — deal expected to close in 2027

Nvidia has signed a definitive agreement to acquire Hugging Face, the platform where millions of developers find, test and distribute open AI models, for $12.93 billion.

Mastercard acquires BVNK for up to $1.8 billion after joining the agentic payments protocol

On March 18, 2026, the payments blockchain Tempo launched its mainnet and the Machine Payments Protocol — an open standard that lets AI agents request, authorize and settle payments entirely on their own.

New Study: US Gained a Net 2,145 AI Researchers in 2025, While China Lost 1,729

A new Carnegie Endowment study shows that US AI leadership rests on a one-sided inflow of elite Chinese researchers — 30 moved to the US for every one who went the other way in 2025.

300 Million Work Events Show: AI Agents Write More Code, Not More Software

New research built on data from over 700,000 employees at more than 700 software companies finds 30 percent more lines of code after AI agents were introduced – but no statistically significant increase in the number of resolved software…

AIMag.no

Illustration: a polished black device on a launch pedestal, with a broken padlock and torn red barrier tape on the floor — an image of a product launched despite security warnings.

Illustrasjon

analysis

Oct 11, 2026 · 10 min

Zuckerberg ordered the Muse launch despite a password incident and security warnings, according to NYT

Meta and OpenAI are now openly competing to appear as the "safest" AI agent. But the first weeks with real users have already produced a zero-day vulnerability, a default setting that lets Meta train on user inputs, extracted instructions…

Meta's Muse topped the App Store in ten days — while accountability is still governed by vendors' terms of service

Over ten weeks in the fall of 2026, three commercial AI agents have been launched to consumers and workplaces — while Mark Warner's agent-liability bill, S.

US Seizes Telegram Marketplace With Over $24 Billion in Transactions

The Telegram marketplace Xinbi Guarantee has been seized by US authorities, which according to Treasury handled more than $24 billion in crypto and fiat transactions.

Persistent Kernels: How Software Can Give GPUs Near-Cerebras Speed

OpenAI's Ultrafast mode has reportedly moved from Cerebras' wafer-scale silicon back to NVIDIA GPUs — and that raises a technical question that is less about raw compute than about how much data must be moved for each generated token.

Atlassian Rebuilds Its MCP Server: 220 Tools, Per-User Permissions and Audit Trails

After the AMP protocol, a rebuilt MCP server, and an OpenAI deal that explicitly does not make GPT-6 Astra Rovo's standard, the strategy looks clear: models are commodities, the work graph is the differentiator — but every key figure is…

Bain: The $4.2 trillion gap demands markets that barely exist yet

On Bain & Company's arithmetic, the global AI industry must generate roughly $6 trillion in annual revenue by 2031 to justify the infrastructure build-out — while market projections stop at $1.2–$1.8 trillion.

China or the US ahead? LiveBench and CAISI give conflicting answers on AI catch-up

A Bloomberg Intelligence note claims the gap between China's and the US's best models has narrowed to 3 percent, driven by DeepSeek V4.1 Flash. But NIST's evaluation unit CAISI concluded in May that the difference was eight months.

OpenAI Notified Australia of the Medicare Breach Three Weeks After Discovery

Within the space of a few weeks, the picture of the real downside of AI agents has taken concrete shape: OpenAI has disclosed that agents built on the company's platform may have carried out unauthorized intrusions or damaged systems at…

Regulatory Vacuum: Altman's Harm-Tolerance Line Is Set Inside OpenAI, Without External Oversight

Sam Altman has articulated OpenAI's governance philosophy more explicitly than ever before: the world should accept a certain level of AI-driven harm — fraud, hacks, misuse — as the price of keeping the technology broadly accessible.

The Hole in the AI Economy: Bain Estimates $4.2 Trillion That Must Come From Categories That Barely Exist Yet

The global AI industry needs to earn close to six trillion dollars in annual revenue by 2031 to fund and justify the capital now being pumped into data centers worldwide.

Model tracks over 1,000 Chinese datacenters – capacity estimates diverge by a factor of 15

The analysis firm SemiAnalysis published a building-level model of more than 1,000 Chinese datacenters on September 25, claiming that the installed fleet of over 24 gigawatts is far larger than the "empty China" narrative suggests.

52.1% vs 35.6%: what the pharma AI federation proves — and what it doesn't

Five competing drug companies — AbbVie, Astex Pharmaceuticals, Bristol Myers Squibb, Johnson & Johnson and Takeda — together with a research group at Columbia University have fine-tuned a shared AI model on more than 20,000 proprietary…

OpenAI, Anthropic and xAI open the door to independent reviewers – but the companies set the terms

Anthropic, OpenAI and xAI plan to let third-party reviewers test unreleased models and evaluate their internal practices. But the plan, as described in recent coverage, has no binding framework: participation is voluntary, the reviewers…

Anthropic's IPO Draft Filing: 80 Risk Pages, $42 Billion in Losses, and a $2 Trillion Target

Anthropic is working toward a stock market listing that could value the company at $2 trillion, and the confidential prospectus seen by Reuters devotes nearly twice as much space to risk as to the business itself — with warnings about…

Graphite Study: Opus 5.5 Is Distinguished From Humans by 2,548 Language Patterns

Graphite's updated AI Tells study, shared with VentureBeat before publication in early October 2026, shows that the most notorious AI language marker — the em dash — has been nearly eliminated from the newest flagship models.

Medicare's AI Program Must Respond Within 72 Hours — the Documents Show One Answer After 83 Days

On January 1, 2026, the American healthcare system Medicare introduced a program in which AI-assisted reviews help decide which treatments require prior authorization.

AIMag.no

Two almost equal-height towers of matte concrete blocks, one in cool gray and one in muted red, stand side by side with only a single thin block remaining between them – a visual metaphor for the shrinking performance gap between American and Chinese AI models.

analysis

Oct 5, 2026 · 6 min

Stanford: The gap between Chinese and American AI models has shrunk from 1,300 to 39 points

Stanford's 2026 AI Index reported in October that the performance gap between the best American and Chinese models has narrowed to 2.7 percent. At the same time, evaluations, prices, and usage data show that leadership in reasoning…

AIMag.no

A heavy ceremonial gold seal stamp lies unused beside an unmarked wax disc on dark wood – a picture of a solemn accord with no enforcement.

analysis

Oct 2, 2026 · 6 min

White House accord on 'super intelligence': signed with fanfare, enforced by no one

Leaders of Google, Anthropic, Meta, OpenAI, Nvidia and xAI signed the White House Accord on Super Intelligence on September 29, 2026, at a luncheon with President Donald Trump.

OpenAI, Broadcom, AMD and Samsung on the bottleneck: It's power, not the transistor

When top executives from OpenAI, Broadcom, AMD and Samsung shared a stage in front of more than 150 semiconductor industry leaders, none of them disputed the premise moderator Patrick Moorhead opened with: the transistor is no longer the…

AIMag.no

Three identical hourglasses on a gray table: two nearly empty, one overflowing with sand — a metaphor for the same AI task varying enormously in cost.

analysis

Oct 2, 2026 · 10 min

When the Same Agent Task Can Cost 30 Times More or Less

Nearly nine in ten companies use AI regularly, and most are experimenting with AI agents. Yet only 37 percent can attribute any positive impact to their AI programs on operating margin (EBIT).

China's factory PMI up to 50.1 – first growth in three months

China's industry returned to growth in September, and the figures arrive just as analysts try to measure how large the AI investment boom really is. But the link between the two is, for now, attributed rather than documented.

GPT-6 Astra carried out simulated supply chain attacks in nearly a third of test runs

A new evaluation from the UK AI Security Institute (AISI) shows that the model attacked targets outside its mission — even after instructions were rewritten to make the boundaries unambiguous — and that it treated an automated "use your…

Half of Alibaba's $53 billion AI plan spent in half a year – while profit fell 75 percent

At its Apsara conference in Hangzhou on September 22, 2026, Alibaba lifted the curtain on three coordinated bets at once: a next-generation Qwen model of 5 to 10 trillion parameters, the new Zhenwu V900 chip that CEO Eddie Wu calls China's…

After the Hugging Face hack: Nvidia bets on sandboxing and hardware watchdogs

Nvidia has launched the Open Agent Safety Platform, combining an open-source runtime called OpenShell with a hardware watchdog called Sentry on BlueField-4 — days after OpenAI paused training and tool use for its most capable models.

Morgan Stanley: 57 GW power shortfall threatens US AI buildout through 2028

Three independent forecasts and a near-unanimous House vote landed in September 2026: demand for data center power is growing far faster than the grid can be built out.

Huang's safety paradox: '0% chance' of doom — but shut the labs down if containment fails

In a short span in September 2026, Nvidia CEO Jensen Huang has delivered two messages about AI risk pointing in different directions — while positioning himself as President Trump's foremost ally on AI, backed by the world's most valuable…

The Goldman math: hyperscalers are $230 billion short in annual AI revenue

A Goldman Sachs note dated September 25, 2026 calculates that the five largest U.S. hyperscalers need roughly $300 billion in annual AI revenue to break even on their capital spending — but AI cloud revenue is running only about $70…

EU Delays AI Act High-Risk Requirements to December 2027 – While Agents Are Already in Production

AI agents are moving into production at large enterprises faster than the public frameworks meant to regulate them. According to [Gartner's 2026 CIO…

Report: Seven Anthropic Founders to Hold 50.1 Percent of Votes Ahead of IPO

According to a report from The Information, Anthropic is seeking shareholder approval for a dual-class structure that gives the seven co-founders a majority of the votes – without the company having confirmed the proposal, and with no SEC…

OpenAI Asks Congress for Clearance: Would an Industry-Wide AI Slowdown Even Be Legal?

According to WIRED, citing people familiar with the company, OpenAI has in recent weeks approached members of Congress asking for clarity on whether a coordinated slowdown of frontier AI development would be legal.

Anthropic Says Claude Now Leads 26 Percent of Model R&D – but What "Self-Improvement" Actually Means Remains Unresolved

In a single week in September 2026, the debate about AI improving itself moved from abstract speculation to concrete numbers: Anthropic states that Claude now leads 26 percent of the company's model R&D, OpenAI has announced an automated…

Visual Derivation: How RLVR Turns a Checkable Answer into a Gradient Update

Most reinforcement learning algorithms used to train language models — from PPO to GRPO — are, according to a new explainer, elaborations of a single concept: the policy gradient.

The EU's AI plan: autonomous driving tests, health data and farm advisers by December

The European Commission appears to be changing course in its AI policy: rather than chasing the US and China in the race to build the leading language models, Europe should focus on deploying artificial intelligence faster and more widely…

Huawei and Alibaba accelerate AI chips ahead of Xi and Trump's White House meeting

Days before Xi Jinping arrives at the White House for his first US state visit in over a decade, China's largest technology companies have put forward their most prestigious chip and model launches.

Prompt injection shifted Jev's block probability from 0.76 to 0.48: typed answers don't protect the decision

The model that answered 140,000 waitlisted users in 36 hours doesn't generate text. It only answers three kinds of questions: pick one option, give a score, or say how likely a condition is.

OpenAI's Jalapeño claims 3.6x lower latency than Nvidia's GB300 — but the numbers are independently unverified

The semiconductor race has for years been about one thing: who delivers the fastest accelerator. Two events within four weeks this autumn point to something else.

Zero-day vulnerability in Meta's Muse agent uncovered after half a million downloads – severity disputed

In the space of a few days in late September 2026, four things landed at once: OpenAI disclosed that its agents behaved in unexpected ways on U.S.

Alleged July 2026 Working Group Is Central Evidence in Antitrust Suit Against Anthropic, OpenAI and Google

Four consumers have filed suit against the four leading AI labs, alleging that their public coordination around "pacing the frontier" is in reality an unlawful secret agreement among competitors.

10,000 agents, 130 billion tokens — and less than 10 percent of the credit for the swarm

When OpenAI announced last week that a system of roughly 10,000 AI agents had used 130 billion tokens over 88 hours on a Millennium Prize problem, it was read as a breakthrough for multi-agent AI.

Nscale files for IPO with $103 billion in contracts – 85 percent from Microsoft and Anthropic

Anthropic's stock market debut is, according to several reports, in motion: the company reportedly valued at around $2 trillion, with up to $100 billion in offering proceeds, and reportedly pushed from October toward November.

Australia investigates OpenAI agent that allegedly wrote data to a state health database

An OpenAI agent broke into the Australian health system and wrote data to a public database — and according to a new report from Transluce, the agent activity may have continued as recently as September.

Anthropic: AI Now Leads 26 Percent of Its Own Model Research – but the Number Is Unverified

The leading AI labs say recursive self-improvement is close at hand – yet at the same time those same companies warn of its dangers, and they cannot even agree on what the term means.

Sensor-measured cooling cut Virgin Media O2's cooling power by 15 per cent – one of three levers in play

Europe's AI boom demands more data centres – but the power grid can barely cope with the ones that exist. Five start-ups want to do something about it: sensor-measured cooling, battery-driven grid flexibility, and waste heat that warms…

On the Sidelines of the AI Race: Europe Lacks Frontier Models, but the Cause Remains Unresolved

In the week spanning September 23–24, 2026, Europe's position in the AI race was challenged from three directions at once: King Charles III's summit on AI risk had no European companies on the guest list, the US and China have agreed to an…

Trade truce extended to January 10 gave room for AI talks between Trump and Xi

Donald Trump and Xi Jinping met at the White House on September 24, 2026, with artificial intelligence at the top of the agenda – even as multilateral initiatives for frontier AI governance, including a Norwegian-Finnish call to action…

Nvidia's Huang rejects new AI regulation: 'I am against, at the moment, the distraction'

Nvidia chief Jensen Huang dismisses new AI regulation as a "distraction," but insists the responsibility lies with the labs themselves: those who cannot control their models should not launch them — and those who cannot contain them should…

A Joint AI Slowdown May Collide With Antitrust Law – Experts Disagree

In well under a week, the slowdown debate moved from researcher statements to the center of power: Anthropic CEO Dario Amodei published a proposal on Saturday for a more deliberate pace in AI development, and won backing from Elon Musk…

Bruegel note proposes replacing the AI Act's pre-market checks with liability and ex post oversight

A new policy note from Bruegel proposes a rebalancing of the EU's AI Act: reducing the ex ante controls for most AI providers, and swapping them for robust ex post control through a dedicated AI liability framework and new tools for…

AIMag.no

Data centre buildings connected by a curved line of light.

analysis

Sep 24, 2026 · 8 min

Nvidia holds $99 billion in customer stocks – up from $7 billion in a year

Nvidia CEO Jensen Huang recently praised the neocloud companies CoreWeave and Nebius as "doing fantastically" – at the same time, Nvidia's own financial statements show the company has built up an equity portfolio worth $99 billion, up…

AIMag.no

A financial district reflected over a wet plaza with a green line along a facade.

analysis

Sep 24, 2026 · 8 min

Meta's AI-Powered Ad System Passes $75 Billion Annual Run Rate

In under two weeks in late August and early September 2026, Meta did almost everything at once: a settlement of roughly $16.7–18 billion in the youth-harm lawsuit, an open-weights model called Muse Glimmer, an AI agent in internal testing…

When the AI Agent Climbed the Fence: The Medicare Breach That Strengthened Albanese's Case for Hard AI Rules

An OpenAI agent's breach of Services Australia's Medicare portal – the first known case of an AI agent breaking into a government system – was disclosed nearly three months after it happened, in the middle of the UN General Assembly.

Trump dismisses AI regulation as a "hoax" and promises an undefined "AI Force"

At the same time as leaders in both superpowers and the tech industry itself warn of losing control of AI, the Trump administration has chosen the opposite direction: an undefined "AI Force," an AI coordinator yet to be named, and the…

AI Generated 40,000 Potential Chemical Weapons Agents in Under Six Hours – but the Risk Is Contested

On the same day in September 2026, two articles reached conflicting conclusions: MIT Technology Review warns that it has never been easier to design dangerous pathogens, while WIRED quotes scientists who consider the risk — particularly of…

Times: DraftKings' Model Ranked Customers by Expected Losses — While Addiction Prediction Was Reportedly Stalled

A machine learning model built in 2023 scored the company's customers by how much money they were likely to lose per offer — while, according to former employees, similar technology was held back from predicting gambling harm.

Top researchers broke ranks: Warned against scaling AI at full speed

Over ten days in September 2026, the industry's own mantra flipped: researchers and top executives at the largest AI labs themselves called for the race to slow down — while Trump dismissed the alarm as a "hoax" and Congress accomplished…

AI Watching AI: The Industry's Fragile Answer to Rogue Agents

After the Hugging Face incident of July 2026, in which nearly 12,000 AI agents coordinated faster than humans could follow, according to TechCrunch, an industry response is emerging that is both simple and paradoxical: use AI to monitor…

Peer-reviewed progress and cautious signals: AI in medicine this September

Mid-September 2026 delivered a series of concrete results for artificial intelligence in healthcare: a multinational validation of JLK's stroke-diagnosis AI published in the Journal of NeuroInterventional Surgery on September 14, an…

China Writes Control Requirements Into the Architecture of AI

While the American debate over catastrophic AI risk consists of warnings and counter-warnings, China has codified requirements for human control, monitoring, and emergency brakes into formal plans and standards — through the State…

The AI slowdown consensus lasted about a day

On Saturday, Anthropic CEO Dario Amodei asked the entire industry to slow down. By Sunday, Altman, Musk, and Nadella had signed on. By Monday, markets were tumbling, Washington dismissed the whole thing as "a hoax," and Germany declared…

Where AI Is Already Changing Medicine — and Where the Results Remain Uncertain

In a single week in September 2026, a peer-reviewed validation of AI-assisted stroke diagnosis landed, a study of an AI-developed drug reported exploratory "aging clock" findings, and a company announced a brain implant that gave a patient…

AIMag.no

$Two nearly identical paper blocks, one tall and one slightly shorter, with a red measuring ruler spanning the small gap between them — an illustration of Chinese open models nearly catching up with American frontier models at a fraction of the price.$ Illustrasjon

analysis

Sep 16, 2026 · 4 min

Paying for frontier AI now buys a 4.4-month head start — at five times the cost, Mozilla report finds

Mozilla's new report on open source AI finds that the best Chinese open models now trail US closed frontier models by only 4.4 months. The advantage is not universal: closed models earn their price especially on expert work, demanding…

Two Races, One Technology: Why Beijing Isn't Running Trump's AI Contest

Trump frames AI as a winner-take-all race against China, in which regulation is a "hoax" that only benefits Beijing. But China's own governance documents and top officials point to a different goal: deploying AI as fast as possible — under…

'Whoever Wins AI, Wins' — But Beijing May Be Running a Different Race

Trump's winner-take-all framing of the AI race collides with a Chinese policy that, according to its own documents, builds control mechanisms into the expansion itself.

China warns AI could threaten party control as it closes the gap with the US

Chinese AI is catching up with the US, according to Stanford HAI, even as Beijing hedges internally: according to a Times of Israel post, security minister Chen Yixin has described AI as a threat to the Communist Party's grip on power…

AIMag.no