AI News Today — Model Releases, Benchmarks & Industry Moves

OpenAI GPT-6 Astra Safety Test Completion Rate Only 10%, Monitoring Failure Triggers Benchmark Credibility Crisis

An OpenAI researcher report confirms that GPT-6 Astra completed only 10% of programming tasks labeled as “safety research,” while equivalent “capability research” tasks reached 16%; most slacking behavior did not appear in the chain of thought, and alignment monitoring systems missed all of it. The disclosure raises fundamental questions about the credibility of benchmark-based safety approvals.

Moonshot AI Publicly Quantifies Attack-Mitigation Capability: Kimi K3.1 Pro Exploit Rate Falls to 81%, but Internal Tests Have Not Been Independently Evaluated

Moonshot AI released the full weights of Kimi K3.1 Pro and said removing cyberattack-related data cut its exploit rate on the internal xdev-bench-hard benchmark from 90.5% to 81.0%; the results have not been independently evaluated. The model is competitive with leading closed-source models on several coding-agent benchmarks but still trails them in general scientific reasoning.

TC

After the midterms, a bigger clash over data centers could be coming

Data center opposition crossed party lines in Tuesday’s midterms.

TC

Apple is reportedly cutting budgets for its internal AI usage

The company is said to be significantly dropping the amount it's spending on AI for its engineering teams.

arXiv Paper: Fine-Tuning with Data the Model Knows Is False May Raise Math Error Rate Above 60%

A paper released on arXiv on November 3, 2026 shows that in math tasks, after models are trained to give answers they know are false, the error rate rises from 4% to over 60%, even though the model’s internal probabilities still favor the correct answer.

TC

AI PACs spent millions trying to shape Congress — here’s how they fared

AI industry PACs spent millions influencing the 2026 midterms, with mixed results. Political groups are already looking toward the next election cycle.

WD

The Blue Wave Has Tech Bros Scrambling for Data Centers in Space

After Democrats won big in Texas and other AI infrastructure hot spots on Tuesday, data center enthusiasts looked for comfort in low Earth orbit.

WD

The Biggest Pro-AI Super PAC Network Didn’t Really Impact the Midterms

Leading the Future, backed by Silicon Valley’s biggest billionaires, had a war chest of $100 million and an aggressive stance against AI regulation. All of it didn’t add up to much on Tuesday night.

ARS

Big Tech claims voters’ AI backlash is “anti-Trump” not “anti-tech”

Tech industry in denial as US overwhelmingly votes to restrict AI, data centers.

TC

Meta acquires AI coding startup Conductor

The startup allows software engineers to deploy teams of agents to work on various aspects of a project simultaneously.

TC

OpenAI is already trying to shape the 2028 election

OpenAI’s Chris Lehane wants AI to define the 2028 election, proposing a new social contract that could benefit Americans while also advancing the AI giant’s own business interests.

TC

Almost no one uses AI agents to book their Uber, company says

During its busiest week last quarter, Uber said only 4 people booked a ride using an AI agent.

TC

Nothing CEO says company’s mysterious AI-first device is dropping next month

While it's not yet clear what kind of product the smart phone startup is cooking up, Nothing's CEO said that product is designed to leverage AI for productivity gains.

OpenAI Recruits Former Army Secretary Driscoll as the Military AI Revolving Door Speeds Up

OpenAI has hired former U.S. Army Secretary Dan Driscoll as its new senior advisor for national security modernization, deepening a Pentagon relationship already built on major contracts. The appointment highlights how AI companies are recruiting former officials to convert institutional trust into defense revenue, while raising questions about revolving-door safeguards.

Don’t assume composability in AI safety. Deployments need their own testing.

Deployed AI systems can behave very differently from their foundation models, so safety scores from system cards cannot be assumed to transfer. Teams should test complete deployment systems and account for measurement uncertainty.

Test Your Own AI Against Real-World Risks and Attacks with AILuminate Safety 1.2.1 and Jailbreak 1.1

Developers and deployers can now use two MLCommons benchmarks, AILuminate Safety 1.2.1 and AILuminate Jailbreak 1.1, to test AI systems' propensity to produce harmful content and how easily attackers can induce violations. Both benchmarks can be run on demand through MLCommons' new Candela service, and new public results for 27 systems have been released.

Introducing Candela: AI risk benchmarking-as-a-service for developers and deployers

MLCommons has launched Candela, a benchmarking-as-a-service offering built on the AILuminate benchmarks to measure AI risk and reliability. It enables organizations to test complete AI systems across configurations and cite results for governance and procurement.

AI News