JadidJust-in AI Developments, Incidents & Deadlines
November 2026
SMTWTFS123456789101112131415161718192021222324252627282930

Everyone running coding agents is now rationing them, and the first hard limits are landing on engineers.

19 stories · transcript

SpaceX cut Grok Bot quotas because serving it cost too much, Cognition added budgets that stop a Devin session mid-task, Goldman Sachs pulled coding tools from everyone but its engineers, and Google counted stolen AI access running up bills of $100,000 a day. Anthropic's amended S-1 shows why vendors are tightening: $17.5 billion of Q3 operating costs against $17 billion of revenue.

TLDR

  • Anthropic's amended S-1 shows Q3 revenue near $17 billion, up 47% on Q2, and a $23.3 billion net loss (Nov 2); compute commitments reach $895 billion. Business Infra
  • Red Hat has patched the CVSSCommon Vulnerability Scoring SystemA standard 0 to 10 severity score: 0.1–3.9 low, 4.0–6.9 medium, 7.0–8.9 high, 9.0–10 critical.FIRST ↗ 9.8 Linux TCP flaw CVE-2026-98432, while Ubuntu fixes sit in opt-in testing (Nov 3). Security Infra
  • Attackers are exploiting Node-RED CVE-2026-58115 (CVSS 10.0), a missing-authentication bug that runs commands (Nov 3). Security
  • SpaceX cut Grok Bot usage limits, calling serving costs "unsustainably high" (Nov 2), and Cognition now ends Devin sessions when a budget runs out. Pricing Agents
  • GitHub moves default blocking of pull_request_target in public repos to Dec 2, with brownouts on Nov 9 and Nov 16 (Nov 2). Security Infra

Upcoming Deadlines

Date What Who it hits Do this Source
Wed Nov 4 Moonshot promises Kimi K3.1 Pro full open weightsA model whose trained weights are published for anyone to download and run.NTIA ↗ (The Decoder, Oct 23) Self-hosters of open models Line up an eval; read the license The Decoder
Wed Nov 4 Apple Intelligence beta with Alibaba Qwen starts in mainland China (iOS 27.3 wide release in December) iOS developers and teams shipping in China Test Siri, Writing Tools and Visual Intelligence flows against the Qwen-backed beta TechNode
Mon Nov 9 SEC roundtable on machine-understandable data and "Super Intelligence" Public-company and data teams Watch for stream SEC
Mon Nov 9 GitHub Actions pull_request_target brownout, 16:00-18:00 UTC: blocked runs fail in public repos Public repos with pull_request_target workflows Check Insights for runs the default rule would block; allow the trigger in an Actions event policy or fix the workflow GitHub
Tue Nov 10 CISACybersecurity and Infrastructure Security AgencyThe U.S. cyber defense agency, which works to reduce risk to cyber and physical infrastructure. It runs KEV.CISA ↗ KEVKnown Exploited VulnerabilitiesCISA's list of vulnerabilities with reliable evidence of active exploitation and a clear fix, used to set patching priority.CISA ↗ federal patch deadline for Brotli CVE-2026-111497 (Node2Shell) Anyone running Node.js, .NET, Bun, Envoy or other servers with a vulnerable Brotli decoder Patch Brotli-dependent runtimes and proxies; confirm transitive copies BleepingComputer
Thu Nov 12 Rust 1.100 stabilizes Cargo minimum publish age (dependency cooldown, off by default; Socket, Oct 20) Rust teams Set a cooldown in Cargo config Socket
~Thu Nov 12 OpenAI's proposed end of model access in Cursor (official date not confirmed; Cursor may end it sooner) Cursor users on OpenAI models Check Cursor terms OpenAI
Mon Nov 16 GitHub Copilot retires Claude Opus 4.8, Claude Opus 5, GPT-5.6 Terra (replacements Opus 5.5, GPT-6.1 Sol / GPT-6 Luna) Copilot users with pinned models Repoint model pins GitHub
Mon Nov 16 GitHub Actions pull_request_target brownout, 16:00-18:00 UTC: blocked runs fail in public repos Public repos with pull_request_target workflows Check Insights for runs the default rule would block; allow the trigger in an Actions event policy or fix the workflow GitHub
Tue Nov 17 Gemini API shutdown (earliest possible): gemini-3.1-flash-tts-preview, gemini-3.1-flash-live-preview, gemini-2.5-flash-native-audio-preview-12-2025, gemini-2.5-flash/pro TTS previews TTS and Live API callers Move to gemini-3.8-live or gemini-3.8-flash-tts Gemini API
Tue Nov 17 Google Gemini Gems become Skills Gems users Review Gems Google

Action Items

  1. Patch every RHEL 8, 9 and 10 host to the fixed kernels from RHSA-2026:93412, 93413 and 93414 and reboot. On Ubuntu 22.04, 24.04 and 26.04, install from -proposed on internet-facing hosts or apply Red Hat's nftables rule that drops inbound IPv4 TCP packets carrying IP options, which is a partial mitigation.
  2. Find every Node-RED instance, including those embedded in Siemens SIMATIC IoT2050 Advanced devices, and take it off the internet or upgrade to Industrial OS 4.3.4.1 or later.
  3. Run the Insights check for pull_request_target in every public repo before the Nov 9 brownout, then either allow the trigger in an Actions event policy or remove fork-code execution from the workflow.
  4. Alert on spend spikes in cloud AI APIs, and rotate any exposed cloud or model keys this week. Google's incident data shows more than $100,000 in charges in a single day.
  5. Pin MCPModel Context ProtocolAn open standard that lets AI apps connect to data sources, tools and workflows through one common interface.MCP ↗ servers to maintained packages in agent configs and ban npx installs of archived ones, starting with the deprecated Anthropic reference servers.

Security

  • Linux TCP flaw CVE-2026-98432: Red Hat ships fixes, Ubuntu's are still opt-in (Nov 3) Security Infra Critical · CVSS 9.8 · Exploitation unknown · Patched in RHEL 8.10 / 9.8 / 10.2 kernels; Ubuntu 22.04, 24.04 and 26.04 in -proposed only Update — Red Hat raised its rating from Moderate 5.9 to Critical 9.8 and shipped errata Nov 1; a reboot is required. Canonical's fixes sit in an opt-in testing channel with no date for the regular one, and Ubuntu 20.04 and older have none. An unauthenticated attacker triggers the race with IPv4 Record Route or Timestamp options. (The Hacker News)

  • Attackers exploit Node-RED CVE-2026-58115, a CVSS 10.0 missing-authentication bug (Nov 3) Security Critical · CVSS 10.0 · Exploited in the wild (VulnCheck, first seen Oct 30) · Patched in SIMATIC IoT2050 Advanced Industrial OS 4.3.4.1 Siemens disclosed the flaw Aug 11. It lets an unauthenticated attacker build a flow with an inject node and an exec node that runs shell commands as root. VulnCheck saw a Linux bot delivered Oct 30, then 15 more attacks from one IP by Nov 1, including persistent webshells. (SecurityWeek | Help Net Security)

  • Nvidia Brev breach may have exposed the SSH keys used to reach customer VMs (Nov 2) Security Infra High · CVSS: not yet assigned (no CVECommon Vulnerabilities and ExposuresA public catalog that gives each known security vulnerability a unique ID, a short description and at least one reference.NIST ↗; scanners will be silent) · Exploitation unknown (Nvidia reports no evidence of misuse) · Keys already rotated by Nvidia Customer notices describe unauthorized activity from Oct 4 to Oct 20. Nvidia replaced the Brev-managed SSH key and updated instances, but did not say how attackers got in or whether workloads were accessed. Brev users should audit login history on those VMs for the 16-day window. (Tom's Hardware)

  • Google threat intelligence: LLMJacking bills reach $1 million, $100,000 in a day (Nov 2) Security Production High · CVSS: not yet assigned (no CVE; scanners will be silent) · Exploited in the wild · No patch; restrict keys and set spend alerts GTIG and Mandiant responded to several 2026 incidents in finance, retail, technology and transportation. One group breaks in; others run "LLMLarge language modelA neural network trained on huge amounts of text that generates text, answers questions and summarizes.Anthropic ↗ Mixers" that resell stolen model access, one at about $60 a month for an unlimited plan. Stolen access turns into spend within hours, unlike cryptojacking. (Google Cloud | Help Net Security)

Models & Tools

  • GitHub moves default pull_request_target blocking to Dec 2, with two brownouts first (Nov 2) Security Infra Enforcement was due Nov 2 and now starts Dec 2 for public repos; private and internal repos are not affected. Blocked runs fail Nov 9 and Nov 16, 16:00-18:00 UTC, then the rule returns to evaluate mode. The trigger runs with base-repo secrets, so fork code under it can steal them. (GitHub)

  • Anthropic's Message Threads cut time to first tokenA small unit of data, such as part of a word, that AI models read and generate. Many AI services price by tokens in and out.NVIDIA ↗ 35% on 500K to 1M token requests (Nov 3) Agents Infra The public beta sends only what changed since the last message instead of resending history. Anthropic says the thread stays encrypted in memory and expires after five minutes idle, and token usage does not change. It works on the Claude platform, AWS, Microsoft Foundry and Google Cloud. The 35% figure is vendor-reported. (The Decoder)

  • Cognition adds hierarchical budgets to Devin, and a session stops when its budget runs out (Nov 2) Agents Pricing Admins assign budgets by department, project and engineer and can approve more capacity. Cognition says the old org limits and per-user tiers stopped working once deployments grew past pilots. Teams should decide who gets the override before an agent halts mid-task. (The New Stack)

  • Alibaba Cloud launches Memory, a long-term memory service for personal agents (Nov 2) Agents Customer AI Memory builds a structured user profile across conversations, with developer-defined rules and profile fields. Alibaba reports 85.15% on HaluMem-medium question answering against 81.71% for Zep, and 56.82 on PersonaMem-v2. Those are company figures, not independently reproduced. (Pandaily)

APIs & Pricing

  • SpaceX throttles Grok Bot usage limits because serving costs were "unsustainably high" (Nov 2) Pricing Agents Some weekly limits jumped from as low as 50% to 100% over Halloween weekend. Musk confirmed it was deliberate and gave no new limits. SpaceX's filing says AI makes up 52% of income and its AI division earned $431 million, so even a profitable agent product is being rationed. (TechCrunch | The Decoder)

Industry & Business

  • Anthropic's amended S-1 shows $17 billion Q3 revenue, slower growth and a $23.3 billion net loss (Nov 2) Business Infra Update to the Oct 29 coverage. Revenue rose about 47% on Q2, down from 144% growth in Q2. Operating costs of $17.5 billion beat revenue, with $5 billion of short-term compute. CNBC says most of the net loss is fair-value changes on convertible notes. Commitments total $895.4 billion. (CNBC | The Information)

  • Nvidia reportedly tells cloud operators to keep Chinese customers under 49% of new business (Nov 3) Business Policy The FT, relayed by Tom's Hardware, says sales teams passed the cap verbally, with no written policy. It reportedly covers new and future contracts worldwide, not existing leases. Nvidia's statement did not confirm the number. Separately, The Information describes smugglers pairing cheap processors with banned HBM so the memory passes export checks. (Tom's Hardware | The Information)

  • Salesforce reportedly pays $2.7 billion for contact-center AI startup Cresta (Nov 2) Business The Information reports Salesforce announced plans to buy the nine-year-old company and agreed to the price, per a person familiar. Salesforce did not disclose terms. It extends a run of acquisitions that includes the reported $375 million Rune deal. (The Information)

  • OpenAI raises bug bounties to $172,000 with bonuses; Google adds a $250,000 Root-in-Cloud tier (Nov 2) Security Business OpenAI more than quadrupled average midpoints for medium, high and critical findings and added a five-tier loyalty program. Google's top reward needs full owner-level control of arbitrary Cloud organizations with no prior access. Both vendors are paying for what AI-assisted attackers now find. (OpenAI | BleepingComputer)

Policy & Safety

  • Anthropic's S-1 says it is approaching "recursive self-improvement" under its Responsible Scaling Policy (Nov 2) Policy Research The filing says using AI to build more capable AI "could also pose a considerable safety risk, particularly if it sharply accelerates the pace of progress." TechCrunch reads this as a warning that the company is near RSI capability; the filing itself says it monitors and mitigates as it approaches the threshold. (TechCrunch)

  • The New York Times sues Brave, alleging it sold scraped content to AI companies (Nov 2) Policy Data The suit names Anthropic, Perplexity, Mistral, xAI and Cohere as customers of Brave's 30-billion-page index and says publishers could not block its crawler. Three Tribune-owned papers filed a similar suit. Brave had not responded; these are allegations. Teams using Brave's search API for grounding should check the supply chain. (TechCrunch)

AI in Production

  • Glow Labs: coding agents install deprecated MCP servers that will never be patched (Nov 3) Agents Security In 98% of sessions with a deprecated server, the agent filled the name from training data. Four high-severity flaws, including Puppeteer server code execution via prompt injectionInput crafted to change an LLM's behavior or output in ways its builders didn't intend.OWASP ↗, have no CVEs or patches. The packages saw over 17.3 million npm downloads since July 2025, so SCA and EDR stay quiet. (Help Net Security)

  • OpenAI details Pando, a Git backend that cut push latency nearly 5x (Nov 2) Infra Production Quarterly monorepo commits will be up over 2,400% on early 2024 by year-end, and reliability had fallen below 90%. Pando skips per-client packfiles and all-branches consistency, serving close to a petabyte a day. OpenAI stayed with Git for compatibility. Figures are the company's. (The New Stack | OpenAI)

  • Goldman Sachs cuts non-engineers off from Claude Code and GitHub Copilot (Nov 2) Production Business Bloomberg, relayed by The Decoder, reports October guidelines leaving the tools to engineers only, after traders and bankers used them for months. Goldman cites productivity and risk management and says the experimentation phase is over. Staff are reportedly frustrated. (The Decoder)

  • Uber's MCP Gateway governs 800-plus servers and 5,000-plus agent tools (Nov 2) Agents Production A registry holds schemas, ownership and enablement, and a proxy turns MCP calls into HTTP, gRPC or TChannel without changing services. Commenters note it resembles API gateways built over the years. The part worth copying is central enablement control. (InfoQ)

Type to search · ↑↓ to move · enter to open