News, analysis and practical guides.

Ten AI models gave sharply different answers in a gender bias audit

A new preprint tested ten AI models on stereotyped phrases and an artificial moral dilemma. Results differed by model and task, limiting broad fairness claims.

Tavus Griffin: what its one-minute video-call result shows

In Tavus-run one-minute calls, 26 of 54 people believed Griffin-Lite was a person. The study, NVIDIA benchmark and limited preview answer different questions.

Barclays’ Claude rollout has activity figures, not outcome data

Barclays reports broad Claude use in customer support, email triage and software development. Its figures show activity and future targets, but leave error rates, productivity and costs undisclosed.

Fake AI policy invitations targeted US experts' cloud accounts

Proofpoint reports a July credential phishing campaign using fake AI policy invitations. Reuters confirmed one recipient received a suspicious message; account compromise is unproven.

QSSR brings AI upscaling to standard PS5, game by game

Sony is adding QSSR to Wolverine and Ghost of Yōtei on standard PS5. Access is game-specific, and sharper reconstruction does not establish faster frame delivery.

AI use is spreading beyond coding. Its intensity is harder to prove

Surveys show wider workplace AI use, while experiments find gains and failures in specific tasks. Usage data cannot yet show that other jobs will match coding's intensity.

Guardian test finds AI image tools removed religious head coverings

The Guardian reports AI image tools removed religious coverings from synthetic people. Published provider rules leave the treatment of those edits unresolved.

California’s AI workplace laws set separate rules for decisions, layoffs and surveillance

California signed three AI-related workplace bills. Layoff-notice and surveillance rules are expected in January 2027; automated decision rules operate in July.

Gemini 4 Argon launches to cyber testers; API access still has no date

Google announced Gemini 4 Argon for trusted cyber testers, with paid API and AI Ultra access planned next but undated. Vals AI's first independent scores lead on some tasks and trail on others; the public API catalog lists no Argon endpoint.

DoorDash’s AI launches span text ordering, Dasher help and restaurant tools

DoorDash's September 30 AI launches cover text ordering, Dasher support, restaurants and office agents. Access ranges from live SMS help to selected-user and corporate betas.

OpenAI and Synopsys plan a chip-design model that can use EDA tools

The companies agreed to develop GPT-Synopsys around chip-design tools and report early customer engagements. The announcement gives no public launch date, price or measured result for the model.

FTC investigation into OpenAI and Anthropic confirmed to AP

AP reports an FTC spokesperson confirmed an AI investigation. CBS says a spokesperson described planned information requests and FTC Act scrutiny; no current order is public.

Tokyo court says a performer's voice can carry publicity rights, but dismisses TikTok deletion claim

A Tokyo court said unauthorized exploitation of a commercially attractive voice can infringe publicity rights. It dismissed Kenjiro Tsuda's TikTok deletion claims after the posts were removed, without deciding whether the clips infringed.

DeepSeek releases DeepGEMM kernels for Huawei Ascend 950

DeepSeek's new Ascend 950 kernel library offers familiar DeepGEMM interfaces and source code. Its prerequisites and self-reported benchmarks set the limits for developers evaluating it.

Meta brings Muse into small-business accounts and tools

Meta has added business accounts and outside tools to Muse. Owners still need to approve publishing, sending and spending, while access and plan limits vary.

OpenAI’s Decisions API: what developers can verify today

OpenAI announced a finite-choice Decisions API in limited preview. Its recap describes classification and routing, while public integration details remain unpublished.

Continues OpenAI DevDay 2026: launches, availability and what builders can use

OpenAI DevDay 2026: launches, availability and what builders can use

OpenAI's DevDay recap lists 20-plus launches across models, Codex, APIs, plugins and teams, from available releases to previews and future rollouts.

Jev AI-judge results vary by task and fallback model

Three preprints test Jev as an AI-answer judge. Results vary across preference, rubric and medical tasks, and confidence routing helps only when its fallback corrects Jev errors.