Ground Truth.
AI, checked against the source.
← 2026-10-032026-10-04later →

Aleph Alpha releases Kolibri, a German-English model with a 78 GB weight footprint

2026-10-04

Aleph Alpha released Kolibri for self-hosted German-English workloads, with sparse computation but substantial GPU memory requirements.

open-weights · sovereign-ai · mixture-of-experts · german · inference

David Robinson leaves OpenAI and argues safety needs more than faster iteration

2026-10-04

David Robinson resigned from OpenAI and said its sprint-driven culture cannot provide the rigor increasingly capable AI requires.

ai-safety · governance · openai · cybersecurity · ai-security · agent-containment

Simon Willison calls for default spending stops as AI agents keep the meter running

2026-10-04

Simon Willison called for default hard spending caps, while provider documentation shows that alerts and enforced usage stops remain distinct controls.

agents · billing · cloud · developer-tools · cost-controls

Sanders, Ocasio-Cortez, and Merkley propose federal restrictions on license-plate surveillance

2026-10-04

The proposed Ban Flock Act would restrict federal access to license-plate surveillance and use funding conditions to pressure state and local agencies.

policy · privacy · surveillance · computer-vision · civil-liberties

Hugging Face disables a GLM-5.3 repository labeled for offensive cyber use

2026-10-04

Hugging Face disabled an offensive-cyber-labeled GLM-5.3 derivative but did not publish a case-specific enforcement rationale or removal date.

cybersecurity · ai-security · red-teaming · model-governance · open-weights

Google moves Gemini Pro out of AI Plus and removes Flash and Pro from free app access

2026-10-04

Google says free personal-account Gemini users will lose Flash and Pro on October 9, while AI Plus will retain Flash but lose Pro on emailed dates.

google · gemini · consumer-ai · subscriptions · product-access

Arizona court orders resentencing after an AI video supplied a victim’s imagined words

2026-10-04

Arizona’s appeals court vacated a manslaughter sentence after a judge relied on generated speech presented in the victim’s voice and likeness.

law · synthetic-media · due-process · voice-cloning · ai-accountability

Benzi reports 391 fixes in 500 coding tasks with a compiler-backed agent

2026-10-04

Benzi’s developers report solving 391 of 500 verified coding tasks with a compiler index and a capped review pass, at $37.33 in model-generation charges.

coding-agents · developer-tools · benchmarks · static-analysis · open-source

Pope Leo renews an artists’ alliance as Anthropic’s moral-status outreach draws scrutiny

2026-10-04

Pope Leo XIV called for an alliance to safeguard human creativity, while Anthropic’s published research continues to treat AI consciousness as unresolved.

ai-ethics · model-welfare · consciousness · creative-ai · anthropic

Oscilloscope Diffusion turns existing movement into AI-stylized video with an approved budget

2026-10-04

Uisato Studio’s hosted Oscilloscope Diffusion reinterprets existing video through AI while retaining source movement as guidance and reserving an approved rendering budget.

creative-ai · video · diffusion · shipping-tools · cost-controls

A false “verified source” label flipped correct answers in an authority-bias study

2026-10-04

An authority-bias preprint reports that false answers attributed to a verified source overturned 45–88% of baseline-correct responses in seven of eight tested models.

research · evaluation · authority-bias · sycophancy · retrieval · ai-reliability

← 2026-10-032026-10-04later →