A preprint counts 21 threats across 9 classes in vibe coding, and the risks grow with the authority the agent is given. Harmless tools chain into attacks, research agents can be steered by fake metrics, and frontier models have few guardrails at the sharp end.
A forensic agent investigates suspect video the way an analyst would, a world model spots fakes it was never trained on, and two vendors claim near-perfect scores on the same audio benchmark. Detection is no longer a single classifier.
In a preprint, shopping agents that reasoned longer grew more confident without getting more accurate. Two more preprints say leaderboard leads and test-time gains may be sampling luck. The fix is to audit, not to spend more.
Lightspark will hold money for AI agents, Razorpay already lets one take payment inside WhatsApp, and a podcast guest says the agent economy will dwarf human commerce. The buyers are asking a different question: who is accountable when the agent spends?
One side points to agents that teach themselves and harnesses that remember their edits. The other side says language has no scoreboard, so there is nothing to improve against. Both sides have evidence.
In a preprint, better-scoring models showed no real gain in merged pull requests, while other papers show passing tests can be luck and agents keep working long after solving the task.
In a preprint, computer-use agents spot harmful patterns across screens but carry them out anyway, while other papers show signed handoffs, tool retrieval and cost controls all have gaps.
Java4Shell allows pre-authentication remote code execution in OpenJDK, while two malicious agent-framework releases on PyPI stole cloud credentials on every Python start.
Astra is the frontier model and Sol the cheaper everyday one, but independent tests show the new models still hallucinate often and trail rivals on real work.
Recursion locked in a Phase 3 trial design with the FDA, argenx widened its use of Cradle's antibody platform, and Kleiner Perkins made the case for the platform shift.
Four launches in a week put OpenAI inside courtrooms, sales teams, allied cyber defence and a national government, while California's attorney general took the company to court over its agents.
Scale AI's 20,000-hour study, a hard look at robot datasets, and four new industrial agents all point the same way: the bottleneck in physical AI is the data, not the model.
Crusoe's trained Qwen model scored 74.1% on held-out FinQA questions, ahead of a 1T-parameter rival, at about $8 per 1,000 questions. SenseNova's 9B model loops 32 layers into 56 computations. Small is not a compromise anymore; it is a strategy.
The chip maker reported record quarterly revenue, raised its growth outlook and announced a deal for the model hub, while its board closed the door on Jensen Huang returning as chief executive.
A preprint audit finds a standard machine-learning screening pipeline produces candidates no better than random, while a simple positional statistic, a generative design workflow and a viral forecasting tool all produce results that survive contact with experiment.
The European Central Bank's November stability review puts ambitious AI earnings expectations, debt-funded infrastructure spending and fragile sovereign markets on the same watchlist.
Hyperbolic has executed the first OTC option on GPU rental prices, and ICE has launched a tool to quantify data center construction risk — the compute market is acquiring the instruments of a commodity market.
Singapore is upgrading PayNow, the Fed is weighing tokenized markets, the Dutch are prototyping an offline digital euro, and Uzbekistan just put its national cards on Apple Pay.
Target reported $27.1 billion in quarterly net sales while J&J Snack Foods, BellRing Brands, La-Z-Boy and Dolby reported results that split between growth and cost pressure.
Filings in a single week show a JPMorgan finance chief handing over, a Ford vice chair retiring, and three more companies changing their top teams at once.
A $250 million round at a $1.5 billion valuation, a $57.5 million data-stack raise and $100 million in debt for AI cooling show private AI money still moving — while an Anthropic IPO would test public markets.
AlphaGenome 2 predicts what variants do in individual cell types, new reference frameworks map GBA1 and the T cell receptor, and a long-read atlas charts the failing human heart isoform by isoform.
Three preprints take on pancreatic cancer from different directions: a protein that organises the tumour's surroundings, a pig model that reproduces the human disease, and a drug pair that hits it in combination. A fourth preprint, on brain cancer, underlines the same lesson about place.
No. 5 Duke lost at home to Illinois and No. 2 Michigan State was held to 18 percent shooting by Tennessee, while No. 1 Florida survived a scare from Notre Dame and Indiana beat No. 20 Arkansas.
Giannis Antetokounmpo meets the Milwaukee Bucks for the first time as a Miami Heat player, insisting it is just another game, while the records tell a starker story: Miami has won five straight and Milwaukee has lost ten in a row.
A $36 million settlement over care for seniors, a 10-year sentence for a $19 million Medicare fraud, enforcement actions against 15 defendants who defrauded veterans, and two more cases from Britain's FCA and the FTC.
The Pan American Health Organization announced the United States has lost its measles elimination status. The governors of Pennsylvania, Massachusetts and Connecticut answered the same day, blaming misinformation and urging vaccination.