Skip to content

Archive

October 6, 2026

24 events

Importance: monitoring New models 1 source

Musubi releases PolicyLM-1.7B for moderation based on natural-language rules

Musubi released PolicyLM-1.7B with open weights. It uses rules written in English to decide whether content belongs to a specified category. According to the company, evaluation takes less than 50 milliseconds, and changing the rules does not require retraining.

Importance: important Google New models ✓ official

Google released EmbeddingGemma 2 for local search across text, images and audio

Google released the open model EmbeddingGemma 2, which converts text, code, images, video and audio into a shared space of numerical vectors. It has 740 million parameters, uses the Apache 2.0 license and can run locally without an API key.

Importance: monitoring Regulation and law 1 source

Arizona appeals court orders sentence reassessment over undue influence of AI video of victim

An appeals court in Arizona has ordered a reassessment of the original sentence of 10.5 years in prison for Gabriel Horcasitas. According to the court, the AI video of the victim unduly influenced the judge and made the sentencing proceedings fundamentally unfair. The conviction remains in place.

Importance: monitoring Anthropic Regulation and law update 1 source

Nathan Lambert after the Black October wave: open weights for Claude Mythos were not the main risk, GLM-5.3 was

After the Black October wave of vulnerabilities (PHP2Shell, GitLab, the Binance incident), Nathan Lambert of AI2 maintains that open weights for the Claude Mythos model would not have been a catastrophe – in his view, that threshold was crossed by GLM-5.3. This builds on his earlier criticism of the debate over banning open models.

New: The Black October wave of vulnerabilities included PHP2Shell, GitLab, and Binance; Nathan Lambert argues that Claude Mythos would not have been a problem as open weights; GLM-5.3 is identified as the model that actually crossed the dangerous threshold; Nathan Lambert shifts away from his previous downplaying of the security risks of open models

Importance: monitoring Tools and apps 1 source

Pinterest introduced Beauty Guides to help users prepare for salon visits

Pinterest introduced Beauty Guides, a feature that, according to the company, turns saved inspiration for hairstyles and nails into information for a salon visit. It offers professional terminology, time estimates, a price range and aftercare requirements.

Importance: monitoring Regulation and law 1 source

Ban Flock Act would prohibit federal agencies from using license plate recognition cameras

Bernie Sanders, Alexandria Ocasio-Cortez and Jeff Merkley introduced the Ban Flock Act. It would ban federal use of ALPR systems, block federal funding for state and local governments using these cameras, and allow lawsuits over rights violations. It is a bill, not an enacted law.

Importance: monitoring Anthropic Tools and apps ✓ official

Elevated error rates for Claude Opus 5.5 are under investigation

The Claude Status page reported an elevated error rate for requests to the Claude Opus 5.5 model. The incident is under investigation; the announcement was published on 6 October at 12:24 UTC.

Importance: important Mistral New models update ✓ 3

Sources disagree on the availability of weights for Mistral Large 4

The WIRED article describes Mistral Large 4 as available with open weights, while other sources describe their release as still planned. The public API preview has safety restrictions; selected partners are testing a version with reduced moderation.

New: The public endpoint has safety restrictions.; The weights are scheduled for release in three weeks, after safety testing is completed.; According to the company, it trained the model on its own infrastructure with 4 000 Nvidia GPUs.; Benchmark results are not yet available.

Importance: monitoring Security 1 source

Fraudsters tricked Arup into paying over 25.6 million USD through a fake video meeting

In 2024, an employee at the Hong Kong branch of Arup transferred over 25.6 million USD after a video meeting featuring impersonations of the chief financial officer and other colleagues. According to the article, the fake likenesses and voices were most likely created using AI.

Part of an overview of multiple AI topics; only this event has been covered.

Importance: monitoring New models 1 source

T-Search released with open weights for multi-step retrieval of evidence passages

The authors released T-Search, a testing environment and a live demo. The model searches for evidence passages in a given corpus over several rounds. According to the authors, it achieves an average Recall@10 of 56.0 in a single run and 61.3 when combining three runs.

Importance: monitoring OpenAI Tools and apps ✓ official

OpenAI monitors recovery of login and administrative APIs on API Platform

According to its status update, OpenAI has implemented a mitigation following increased error rates for login and administrative APIs on API Platform. It continues to monitor recovery; login remains marked as degraded.

Importance: important Anthropic Security update ✓ official

Anthropic opens access to Claude models with lighter filtering to individual security researchers

The expanded Cyber Verification Program allows vetted individuals to obtain Defense Access to Claude models with lighter safety filtering. Red Team Access remains limited to organizations; Anthropic vets Specialized Access applicants jointly with the US government.

New: The access allows the use of models with reduced safety filtering.; Defense Access is also available to individual security researchers.; Red Team Access is reserved exclusively for organizations.; Anthropic vets Specialized Access applicants jointly with the US government.

Importance: important OpenAI Coding update ✓ official

The Decisions API costs 0.10 USD per million input tokens, and output tokens are free

The Decisions API from OpenAI costs 0.10 USD per million input tokens; output tokens are free. In public beta, it evaluates text and images using the gpt-6-luna model and returns yes/no probabilities, categories, or ratings on a scale.

New: The price is 0.10 USD per million input tokens.; Output tokens are free.; The API offers yes/no probabilities, category selection, and ratings on a scale.; The API supports zero data retention.; The API supports HIPAA-compliant use in the USA and Europe.