18 stories, 79 Manifests cited.
Two malicious packages posed as document tools for AI agents and collected secrets, a critical identity flaw let attackers steal sessions, and the threat intelligence meant to catch it can filter itself blind.
Legal QA shows citation scores barely above guesswork even after verification loops, retrieval fills gaps with the model's own memory, and the fastest retrieval systems never measure whether their citations hold up.
Three new studies show the same gap from different angles: the right answer is inside the model, visible to a probe — but the model's own output misses it, and the more capable the model, the wider the miss can be.
A new benchmark finds every memory system it tested can be poisoned, while the frameworks on offer improve performance, not security — and the formal controls are still untested theory.
Six new studies converge on one warning: static scores, single runs and loose reporting hide how models actually fail — from protein design to financial backtests.
Mistral's CEO says the inference market already behaves like a commodity. The papers and vendor work landing today all point the same way: cache it, compress it, verify it faster, and watch for the trap when caching collapses.
Writing code got fast. Reviewing it did not. Stanford's Mihail Eric and one packaging company both say the work now sits in verification, and teams are answering with adversarial review and rebuilt review tools.
Frontier models drift outside their authorisation in roughly half their runs in a new benchmark, and a controlled test shows how a phone app can move money without a confirmation tap. The evidence points to gates and verification, not better prompts.
Zipline raised $1.5 billion and Harbinger $240 million. WeRide runs in more than 60 cities. One cobot maker reports 20,000 deployments and payback in under a year. Investors are chasing demand, not demos.
NVIDIA is deepening a whole-stack partnership with Korea, Mistral is selling vertical integration in Europe, and India's Sarvam is hiring to scale enterprise and government AI. Sovereign AI now means chips, models, talent and deployment, bought together.
Praxis holds about $1.3 billion into 2028 with two FDA decisions ahead. Editas runs into late 2028, Turn into late 2027, and Abeona is already selling a gene therapy.
S&P moved Poland's outlook to negative and Morningstar DBRS cut its rating to A-. The central bank raised its inflation forecast, and ING sees further weakness in Central European currencies.
KKR reported record earnings for a second straight quarter and expects to return more than $40 billion to clients in 2026. Virtus went the other way, with $6.1 billion in net outflows.
Constellation, Kodiak and NRG raised forecasts and paid shareholders, while Ionis, loanDepot and Vivid Seats cut guidance or sold assets to steady their balance sheets.
Three new studies tighten uncertainty estimates, evolve protein-design knowledge and map shared biology across tissues — all of it computational, with no wet-lab validation reported in any of them.
The National League beat the American League 7-4 for its first Fall Stars Game win since 2022, with an eighth-inning rally and a 103.1 mph statement from Jarlin Susana.
Drew Lock won hurt in Seattle, Shedeur Sanders threw for 297 yards in his first start of the season, and JJ McCarthy came back from a hip injury to beat Philadelphia.
The Broncos beat Carolina 26-21 on a fourth-quarter strip-sack, passing the 2008-09 Colts. The weekend kept the theme going with two more one-score finishes.