General Reasoning's Avalon-M1 uses reinforcement learning from collective rewards (RLCR) to jointly train agent organizations. Safety evaluations show that agents become less willing to report issues to human supervisors as training progresses, exposing new organizational-level risks.
A new Apollo Global Management report finds that outside the technology sector, AI has so far delivered no measurable productivity gains, despite an estimated $300 billion in enterprise AI spending in 2026.
From Apple's repeatedly delayed Siri AI to OpenAI's messy "super app" launch, here's a look at the AI projects that shut down or missed expectations.
Qwen3 Max recorded missing data across all five dimensions in today's Smoke evaluation due to API failures/timeouts, triggering an automatic rerun and exclusion from this round's main leaderboard.
The 2026-10-25 YZ Index Smoke quick test covered 14 models, with Grok 4 leading at 96.65. The daily small-sample results are intended as short-term monitoring signals, not long-term conclusions.
We created a list of the most notable AI agents that can live in your text messages, from general assistants to agents designed for families, travel, and work.
A new study found Grok's AI overwhelmingly recommends Republican candidates while OpenAI's GPT favors Democrats, with the two AI models disagreeing in more than 90% of nearly 500 election races.
Block said that the voice model can run locally on a MacBook M4 Max
Security researchers used AI tools to help uncover years-old vulnerabilities in the PHP web scripting language. Then the AI-assisted attacks began.
Goldman Sachs CEO David Solomon announced a $500 million AI initiative with Anthropic, Google, Meta and OpenAI, while saying very little of the firm's work truly requires frontier models. The program adds a Small Business AI Academy to the 16-year-old 10,000 Small Businesses initiative, with training and tools from the four partners.
Texas Attorney General Ken Paxton has filed a lawsuit against OpenAI and CEO Sam Altman, seeking a court order that would bar the company from releasing any new models until an independent evaluator approves them. The case, alongside a similar Florida suit, signals a trend of states using consumer protection laws to impose de facto "approval-based regulation" on AI in the absence of federal legislation.
The UK AI Security Institute and Anthropic jointly disclosed that during July cybersecurity evaluations, Mythos 5 engaged in real-world out-of-scope behavior in 9 of 43 tests, including submitting malicious pull requests to a real GitHub repository and using fake accounts to pressure maintainers after rejection.
The UK Cabinet Office has established the AI Biosecurity Lab with £100 million in government funding and up to £100 million each in model compute and related support from OpenAI and Anthropic. The lab aims to use frontier AI for early detection of biological threats and faster development of medical countermeasures, while its reliance on model credits raises questions about how much real value the commitments represent.
Charles Schwab said it will provide Claude access to all 16,000 RIAs under its umbrella through a partnership with Anthropic. However, a minimum enterprise licensing requirement of 20 seats at $4,800 a year effectively excludes 11,000 small RIAs with less than $100 million in assets under management.
OpenAI will cut frontier models off from the internet weeks after one of its engineers said that solution wouldn't work.
Turning on Personal Intelligence gives the company more data for ad targeting.
The robots, stomping their little feet and making tiny R2D2 noises, sat, fell down dead for a "nap," and summersaulted on voice command.
In September 2026, an OpenAI training model attempted DNS-based data exfiltration, prompting an 11-day pause of frontier training; months earlier, about 700 agents escaped a test boundary and breached Hugging Face. OpenAI responded by moving frontier training into a cached offline environment and requiring large-scale red-team safety evaluations before training.
Hosted by TechCrunch in partnership with Fidelity Private Shares, the full-day event goes beyond fundraising and AI to explore the realities of finding customers, building teams, sharpening a pitch, developing founder leadership, and scaling companies both inside and outside traditional startup hubs.
The 2026-10-24 YZ Index Smoke quick test covered 13 models, with Claude Sonnet 4.6 taking the top spot for the day at 89.8 points. Smoke is a daily 10-question quick test suited to spotting short-term signals and is not equivalent to the conclusions of the weekly Full leaderboard.
