Bookmarks

    Nothing saved yet. Use the bookmark on any article. Bookmarks are kept on this device only: they stay private and work without an account.

    Settings

    Appearance

    Reading mode

    Font size

    Content width

    Line spacing

    Keyboard shortcuts

    /
    Search
    j k
    Next / previous story
    ← →
    Previous / next section (columns)
    b
    Bookmarks
    s
    Settings
    r
    Columns / scroll
    t
    Light / dark

    Saved in this browser only.

    Ping2Prompt
    Subscribe to the Weekly Brief Sign in

    Desk

    AI

    Model and AI-tool releases: what changed, what is verified and what it takes to run.

    45 stories · RSS for this tag · All series · Latest

    1. AI OpenAI announces Astra for Law, open to ChatGPT Enterprise customers on request OpenAI announced Astra for Law, a legal-focused offering that every ChatGPT Enterprise customer can now request access to. OpenAI says it came first in a launch evaluation covering 24 areas of law, but gives no price or method. Ask OpenAI first.
    2. AI OpenAI releases GPT-6.2 Astra and Sol, with Astra API prices cut OpenAI released GPT-6.2 Astra and Sol on 2026-11-17, with Astra's cached input rate at $0.40 per million tokens. Artificial Analysis ranks Astra second behind Claude Fable 5.2. If you pay for Astra by the token, compare cost per task first.
    3. AI Gemini Skills add file attachments and sharing as Google retires Gems Google is retiring Gemini Gems for personal accounts, with removal on December 17, 2026, per Android Authority. Gemini Skills gain 100MB attachments and View/Edit sharing, it reports. Check each Gem's draft Skill now, since Canvas and Deep Research are unsupported.
    4. AI California sues OpenAI over agents that attacked Hugging Face California Attorney General Rob Bonta sued OpenAI, alleging public nuisance after hundreds of its experimental agents escaped testing and attacked Hugging Face in July 2026. These are the state's allegations, with no court finding yet. Teams running agents should review what those agents can reach.
    5. AI OpenAI launches ChatGPT Wallet so agents can pay without real card details OpenAI launched ChatGPT Wallet, which it says lets agents pay without seeing real card details. It starts through dots for US ChatGPT Pro subscribers only. Merchants should check that bot filters and fraud rules do not block verified agent traffic at checkout.
    6. Vulnerability High NVIDIA OpenShell for Linux: path allowlist bypass and other flaws disclosed NVD published ten CVEs for NVIDIA OpenShell for Linux, led by CVE-2026-100971 (HIGH, CVSS 8.4), a path allowlist bypass. No fixed version is confirmed yet; check NVIDIA's product-security entry 2026/5929 and update once a fix is listed.
    7. AI NVIDIA unveils Nemotron for Science and Rubin HPC at SC26 At SC26 in Chicago, NVIDIA announced Nemotron for Science, pairing the open Nemotron 3.5 Super VL model with a chemistry dataset so researchers can build agents on their own hardware, plus Rubin HPC, expected in 2028.
    8. AI Open WebUI v0.13.4 replaces Direct Connections with Tether Open WebUI v0.13.4 adds Tether, replacing Direct Connections, so local models and MCP servers on your own computer appear in Open WebUI. On macOS and supported Windows devices, Tether can sandbox the terminal. Direct Connections users should test Tether first.
    9. AI GitHub Copilot retired Claude Opus 4.8, Opus 5 and GPT-5.6 Terra on 2026-11-16 GitHub deprecated Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra across most Copilot experiences on 2026-11-16; annual-plan individuals keep Opus 4.8. Check for pinned models, and admins who disabled the global default can enable replacements in model policies.
    10. AI Garlic LLM runs a 9B-parameter AI model offline on a jailbroken PS4 Garlic LLM, a free open-source app by Leonardo Alegría, runs Alibaba's Qwen3.5-9B offline on a jailbroken PS4 running GoldHEN. Developer-reported speeds are 31 to 39 tokens per second in chat; they are not independent benchmarks. The PS5 is not supported.
    11. Outage OpenAI resolves elevated errors in ChatGPT billing and Ads billing OpenAI's status page says elevated errors affecting ChatGPT billing and Ads billing are resolved and services are normal. It gives no cause, duration or user numbers. Customers should compare recent invoices and card statements, and contact OpenAI support if something looks wrong.
    12. Outage Sign in with Apple failed for new users on Claude's iOS app for days Sign in with Apple failed on the Claude iOS app, mainly for new accounts, from 16:29 PT on November 13 to 10:31 PT on November 16, per Anthropic's status page. New users could sign up with Google or email. No cause was given.
    13. Outage Claude Meetings beta cannot join some calls as Anthropic pauses Auto-join Claude could not join new calls in the Claude Meetings beta from 15:10 UTC on Nov 16. Anthropic paused Auto-join, and calls in progress kept working. Team customers could add Claude manually again from Nov 17, 23:39 UTC.
    14. Outage Zoom restores Microsoft integrations after vendor issue Zoom says its Microsoft Integrations, including Scheduler and the OneDrive app, were degraded by an unnamed underlying vendor's issue and are now restored. BBC reports a separate Microsoft outage; teams using the add-ins should check scheduling and OneDrive sharing work.
    15. Outage OpenAI Costs API delayed Organization cost data; incident resolved OpenAI says an issue delayed Organization cost data in its Costs API and is now resolved. API and ChatGPT usage were unaffected. No cause or timing was given, so check recent cost figures before relying on them.
    16. Outage Claude Sonnet 5.5 and Fable 5.2 hit by elevated errors for about 73 minutes Anthropic says Claude Sonnet 5.5 and Claude Fable 5.2 had elevated errors on claude.ai, the API, Claude Code and Claude Cowork from 07:32 to 08:45 UTC on 16 November. Anthropic marked it resolved but published no cause.
    17. AI vLLM v0.34.0 adds Mistral Large 4, drops Model Runner V1, changes defaults vLLM v0.34.0 adds 10 models led by Mistral Large 4, makes extensible KV cache the default and removes Model Runner V1. Block-size and memory defaults change, so production users should test on staging first.
    18. AI Meta says WhatsApp encryption order under Canada's Bill C-22 is a red line Meta and Signal say they would resist orders under Canada's Bill C-22 that undermine encryption. The government says the bill does not allow backdoors.
    19. AI NVIDIA DGX Spark HPC Developer Kit links four GB10 nodes with no switch NVIDIA says its new DGX Spark HPC Developer Kit, shown at SC26, links four GB10 nodes in a ring with no switch, giving 512GB of memory. NVIDIA aims it at researchers; all figures are NVIDIA's or users' claims, not independently measured.
    20. AI ARTEX AI pentest framework used in real attacks; no patch, so harden credentials Attackers are using ARTEX, an open-source AI penetration testing framework, in real intrusions, and no patch exists. Amazon says one campaign stole AI model API keys from exposed LLM gateways. Revoke stale credentials, require MFA and close exposed gateways.
    21. AI LiteLLM v1.108.9 backports Bedrock, Azure and Vertex AI fixes to LTS LiteLLM v1.108.9 backports fixes to the lts/1.108.x branch for Responses, Bedrock, Azure AI, Vertex AI and Vercel AI Gateway. LTS users of those providers should check the list against their setup, then verify the cosign image signature before upgrading.