Agent-built software is going live with nobody checking who can reach it, and the bills are now countable.
Claroc found 35,578 Grok-built apps open to anyone, Railway puts $740,000 of fraudulent AI usage on stolen customer keys, and an SGLang flaw let an attacker into NEAR AI's confidential inferenceRunning a trained model on new input to get an output, as opposed to training it.NVIDIA ↗. Meanwhile the vendors kept pushing reach: Anthropic moved into meetings and spreadsheets, Google made Gemini 4 Carbon generally available, and Microsoft showed that fewer than 10 edited neurons can turn a model into a sleeper agent.
TLDR
- Google makes Gemini 4 Carbon generally available at $2/$10 per million tokensA 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 ↗, doubling after Nov 30, with a /swarm multi-agent preview in Antigravity (Nov 10). Agents Pricing
- Anthropic launches Claude Meetings, Whiteboards and Sheets, and Microsoft shares fall (Nov 10). Agents Business
- Microsoft Research shows a model can be backdoored by editing fewer than 10 neurons, and the result still passes evaluations (Nov 10). Security Agents
- Claroc finds 35,578 of 37,571 Grok-built apps open without sign-in, including SpaceXSI's own sales pipeline (Nov 10). Security Agents
- Anthropic's amended S-1 adds about $10B of cloud commitments and says a $34.9B Lambda lease may be cancelled, ahead of an up to $88.5B raise (Nov 10). Business
Upcoming Deadlines
| Date | What | Who it hits | Do this | Source |
|---|---|---|---|---|
| Wed Nov 11 | Node.js security releases for 26.x, 24.x and 22.x (highest severity HIGH) | Anyone running Node.js 22, 24 or 26 in production | Reserve a patch window and upgrade the day the release lands | Node.js |
| 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 | |
| ~Thu Nov 19 | PostgreSQL 19 expected release (delayed, 8-10 features moved to 20; date not confirmed) | Postgres upgrade planners | Re-plan upgrade window | The Register |
| Fri Nov 20 | Meta updated commercial terms take effect (moved from Oct 30 after publisher objections; crawling clause now limited to advertising and business products) | Anyone with a Meta business/ads account tied to their domains | Re-read live terms; check robots.txt and staging logins | TechCrunch · Meta |
| Fri Nov 20 | Conductor Cloud shuts down after Meta acquisition (desktop app stops receiving new models in early 2027) | Teams running parallel coding agents on Conductor | Export work and move to another agent orchestrator | The Decoder |
| Fri Nov 20 | 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 SonicWall SonicOS SSL-VPN SAML bypass CVE-2026-19868 (exploited) | Anyone running Gen7 SonicWall firewalls (TZ, NSa, NSsp, NSv) with SAML SSL-VPN login | Upgrade SonicOS to the fixed release; review SSL-VPN logins for the SAML path | BleepingComputer |
Action Items
- Audit every app your team built with Grok Build mode or any "publish" button, and unpublish or add sign-in today: Claroc says the default publishing setting leaves apps open to anyone with the link, and about 1 in 10 open apps held internal or personal data.
- Patch Xen Orchestra and XOA now and restrict the management interface to trusted networks: VSA-2026-062 is exploited, has 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 ↗ yet, and a crafted cookie yields admin and root on XCP-ng hosts.
- Upgrade SGLang and rotate keys on any inference host that ran it, including inside a TEE: NEAR AI says the TEE did not protect against a flaw in the software running inside it.
- Move Git to 2.56.1 (or the matching 2.50 to 2.55 fix) before the next recursive clone of untrusted repositories, and update apps that bundle their own Git (CVE-2026-111760).
- Pull AI provider keys out of PaaS environment variables, set provider spend caps, and rotate any key stored on Railway: stolen keys there produced an estimated $740,000 in fraudulent usage.
Models & Tools
- Gemini 4 Carbon is generally available, with /swarm in Antigravity (Nov 10) Agents Pricing. Google positions it above Gemini 4 Argon for multi-package migrations and long-horizon agent work. It costs $2/$10 per million tokens, matching Argon, then $4/$20 after Nov 30. First-place benchmark claims are Google's own, and Carbon uses more tokens than Argon. Copilot added it the same day, on by default unless an admin disabled the global default. (The New Stack | GitHub)
- Anthropic turns Claude into a workplace suite (Nov 10) Agents Business. Claude Meetings joins a call as a visible participant, drafts plans and answers in chat while people talk, and files notes in claude.ai. Whiteboards and Sheets launched with it. The Information reports Microsoft shares fell on the news. (TechCrunch | The Information)
- Fable 5.2 completes 41.3% of Remote Labor Index freelance projects (Nov 10) Research Agents. Scale Labs and the Center for AI Safety's index measures whether work is good enough for a paying client. The top system scored 2.5% a year ago and Fable 5 about 16% in July. (The Decoder)
- Google finds stale TODO comments derail coding agents (Nov 10) Agents Research. Two Google engineers show agents following TODO(bug/...) comments for long-resolved bugs into unrequested side-quests, which costs tokens and time and can corrupt logic. Prune dead TODOs before pointing an agent at a legacy repo. (InfoQ)
APIs & Pricing
- OpenAI schedules
gpt-image-2for removal on Feb 10, 2027 (Nov 10) Pricing. Thegpt-image-2-2026-04-21snapshot goes with it; OpenAI points togpt-image-2.6-sunburstorgpt-image-2.6-flare. (OpenAI)
Security
- Microsoft's Lazarus technique backdoors an agent model by editing fewer than 10 neurons (Nov 10) Security Agents. Medium · 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 ↗: not applicable (research) · Not exploited, per Microsoft · No patch; verify weight provenance. The edited neurons rarely fire and fire on a trigger string, forcing an attacker's command into the output, roughly one neuron per token. The model passes evaluations, and the attacker needs write access to the weights, so treat third-party and internally fine-tunedTraining an already trained model further on extra data so it fits a specific task, domain or style.Anthropic ↗ weights as supply-chain artifacts. Scanners will not flag this. (Microsoft | Help Net Security)
- Grok-built apps expose business and patient data (Nov 10) Security Agents. High · CVSS: not applicable (no CVE) · Exploitation unknown · Remediation: unpublish or add sign-in. Claroc found 35,578 of 37,571 apps open, including 24 built by X and SpaceXSI staff, which have been taken down. About 2% of the open apps held high or critical material, such as a hospital group's list of more than 3,500 employees. Claroc blames the default publish setting, guessable addresses and weak access control in generated code. (The Hacker News | The Decoder)
- SGLang remote code executionRemote code executionA flaw that lets an attacker run malicious code on a target's computers or network.Cloudflare ↗ breaches NEAR AI's inference (Nov 10) Security Infra. Critical · CVSS: not yet assigned in the report · Exploited in the wild · Fix: update SGLang. The attacker ran code on a server inside a trusted execution environment, which the company says cannot protect against a flaw in software running within it. About 100 Private Chat users' account details and 18 agents on an older deployment were reached. (The Hacker News)
- Railway confirms $740,000 of fraudulent AI usage from stolen environment variables (Nov 10) Security Infra. High · CVSS: not assigned · Exploited in the wild · Remediation: rotate secrets. Update — The attacker decrypted variables in at most 1.85% of workspaces, about 100,000, and queried for AI provider keys. Railway says it will cover charges an AI provider refuses to refund. (The Hacker News | BleepingComputer)
- Xen Orchestra authentication bypass VSA-2026-062 is exploited (Nov 10) Security Infra. Critical · CVSS: not yet assigned · Exploited in the wild · Vates has issued a fix; apply the latest release. A cookie of
authenticationToken=j:%7B%7Dmakes the REST layer return the first stored token and log in as its owner, then reach root on XCP-ng nodes. No CVE means scanners may be silent. (BleepingComputer | The Hacker News) - Git fixes code execution during recursive clones (CVE-2026-111760) (Nov 10) Security Open Source. High · CVSS: not yet assigned · Exploitation unknown · Fixed in Git 2.56.1 and 2.50.2 through 2.55.1. An integer truncation lets a long path land as
.gitin a submodule, replacing it with a symlink to attacker hooks. Apps that bundle Git need their own update. (The Hacker News) - Patch Tuesday volume stays above 1,000 flaws, with one exploited Outlook spoofing bug (Nov 10) Security. Medium · CVSS 6.5 · Exploited in the wild · Patched in the November updates. CVE-2026-124194 lets a crafted message file make Outlook connect to an attacker-controlled location without the usual warning, which can leak NTLMv2 hashes. Outlets count 1,235 or 1,259 flaws depending on whether non-Microsoft CVEs are included. (The Hacker News | BleepingComputer)
- Fortinet, Ivanti and Citrix patch batches include criticals (Nov 10) Security. Critical · CVSS 9.8 · Exploitation unknown · Fixed in EPM 2024 SU8 and CSA 5.0.8.1. Ivanti's EPM SQL injection (CVE-2026-121653) allows unauthenticated code execution, and its CSA flaw (CVE-2026-121683, CVSS 9.4) is a missing authentication issue. Fortinet published 89 advisories, about 17 critical, three weeks after it began AI-assisted product testing. (SecurityWeek | SecurityWeek)
- N-able ships hotfixes for the N-central zero-daysAn attack using a flaw that was unknown before it was exploited.NIST ↗ (Nov 11) Security. Critical · CVSS: not yet assigned · Not exploited evidence, per N-able · Hotfixes released. Update — The hosted environments taken offline Monday after the public disclosure are covered by the fix, and on-premises customers need to apply it. (BleepingComputer)
Industry & Business
- Anthropic's amended S-1 adds roughly $10B in cloud commitments and flags a possible Lambda cancellation (Nov 10) Business. Update — New Microsoft and Google commitments run through 2033, and a $34.9B Lambda lease may be cancelled. Crunchbase counts up to $88.5B raised across the IPO and a $26B private placement, the largest ever tied to a listing. (The Information | Crunchbase News)
- OpenAI's Q3 revenue reaches $10B, with operating losses still growing (Nov 10) Business. A person familiar told The Information revenue rose nearly 50% quarter over quarter while the adjusted operating loss grew from $10.1B to $10.9B. The Decoder sets that beside Anthropic's roughly $17B of revenue in the same quarter. Both figures are unaudited and reported secondhand. (The Information | The Decoder)
- CoreWeave beats and raises, easing the neocloud debt scare (Nov 11) Business Infra. Revenue rose 168% to $3.66B and the net loss was $709M, with about $9B in customer prepayments and deposits. Shares jumped about 15% premarket after its credit spreads had passed 11%. SoftBank wrote off $3.7B of non-refundable data center advance payments the same day. (CNBC | The Information)
- Nvidia's board hires Wachtell for the Huang investigation; Congress sets a deadline (Nov 11) Business Policy. Update — The firm, led by former SDNY acting U.S. Attorney Joon H. Kim, will examine whether company secrets or national security were compromised. House Foreign Affairs Chair Brian Mast says he will subpoena Huang if he does not respond today to a Nov 18 briefing request. Nvidia reports on Nov 17. (CNBC | Tom's Hardware)
Policy & Safety
- New York opens its RAISE Act registration portal (Nov 10) Policy. Frontier developers with more than $500M in annual revenue must register by Jan 1, 2027, publish safety frameworks and report critical incidents within 72 hours. Penalties reach $1M for a first violation. (TechCrunch)
AI in Production
- A Linux memory-management reviewer steps back, citing AI patch volume (Nov 11) Open Source Production. Lorenzo Stoakes is leaving core, reclaim, Rust and THP review because the time spent on AI-written submissions is, in his words, too much. The load on maintainers is the cost behind the fix volume Torvalds praised yesterday. (Tom's Hardware)