Doyensec has open-sourced ElectroNG: ~60 Electron security checks, 650+ known vulnerabilities, SARIF export. Important caveat: the authors say major development stopped in 2024, so modern Electron coverage may be noisy or incomplete. blog.doyensec.com/2026/10/27/e...

ElectroNG is now open source!

Doyensec releases the Electron security scanner and analysis engine under Apache-2.0, with current-coverage caveats.

blog.doyensec.com

Anthropic says an unreleased Claude model chained bugs across Chrome, Safari, Android and Linux. Key risk: separate flaws combine to cross security boundaries. The company says it disclosed findings to vendors. www.anthropic.com/research/dis...

Discovering and exploiting vulnerabilities in critical, real-world software with Claude

Anthropic Frontier Red Team's report on chained vulnerabilities in browsers and operating systems.

anthropic.com

pg_pinecone (Apache-2.0) runs IVF+RaBitQ vector search inside Postgres, no Pinecone account required. Filters run during search, not after HNSW retrieval. Pinecone reports a 136 MB index for 1M×768d vectors; benchmark is vendor-run. www.pinecone.io/blog/introdu...

Introducing pg_pinecone: Open-Source Search for Your Postgres Data

Apache-2.0 PostgreSQL extension with native RaBitQ vector search and filter-aware retrieval.

pinecone.io

Replit Cloud Agents can turn a PM's request into a PR with sandbox browser tests, screenshots and video. Engineers still control merges. Review shifts from “what changed?” to “what did the app do?” A demo isn't a regression suite. replit.com/blog/anyone-...

Anyone can now bring their ideas into the company codebase

Replit Cloud Agents build and test code changes, attach video evidence to pull requests, and leave final merge decisions to engineers.

replit.com

Meta's PrivacyLib puts privacy checks inside data-access clients. Compiled policies run locally; agent tool calls carry session context to the data boundary. This scales, but only if bypass paths are covered. engineering.fb.com/2026/10/27/s...

PrivacyLib: Privacy Controls at the Data Access Boundary

Meta's shared privacy enforcement library checks data access in clients and propagates agent context.

engineering.fb.com

GitHub Copilot SDK can now authenticate org-run agents with short-lived GitHub App tokens. No personal login; the org owns billing, model policy and budgets. Token refresh becomes part of agent ops. github.blog/changelog/20...

GitHub Copilot supports server-to-server authentication

Organization-owned GitHub Apps can authenticate Copilot SDK agents without personal user credentials.

github.blog

LiteLLM piloted private patches for severe flaws before disclosure. Parallel releases were costly; getting affected customers to upgrade was harder. Embargoes only buy safety when patch delivery works. docs.litellm.ai/blog/litellm...

Introducing the LiteLLM Private Registry

Private security fixes, release-engineering costs, and patch adoption from LiteLLM's October 27 report.

docs.litellm.ai

Bito's Governor routes coding-agent tasks by risk, preserves model choice mid-task, and accounts for warm-cache cost. Bito reports 48% lower cost/task in one customer A/B test, not an independent benchmark. bito.ai/blog/governo...

Introducing Governor, a context-driven model router for coding agents

Governor routes coding-agent tasks by risk and code context, with vendor-reported reductions in cost per task.

bito.ai

Cloudflare reports 155 GLM-5.3 responses exposed other users' data. An oversized prompt was rejected at prefill, but SGLang decode read stale KV pages from other requests. KV ownership is a security boundary. blog.cloudflare.com/kv-cache-exp...

Cross-request KV cache exposure in the GLM-5.3 Workers AI model, and our remediation with SGLang

A configuration mismatch in GLM 5.3 on Workers AI triggered an SGLang bug that exposed KV cache pages across requests.

blog.cloudflare.com

Akamai Agent Sandbox runs agent-written JS/Python in a restricted runtime nested inside Spin's WASM sandbox. Egress is blocked by default. A real boundary around generated code matters more than prompt-level rules. Beta. www.akamai.com/products/aka...

Akamai Agent Sandbox | Akamai

Run AI agents and the code they write in a WebAssembly sandbox on Akamai Functions, with outbound access blocked by default.

akamai.com

Qt disclosed CVE-2026-105946: NTLM authentication in Qt Network can read out of bounds when a malicious server/proxy requests NTLM. Patch to 6.8.10, 6.11.2 or 6.12.0. Client-side auth parsing is an attack surface too. www.qt.io/blog/securit...

Security advisory: CVE-2026-105946 impacts Qt NTLM authentication

Out-of-bounds read in Qt NTLM authentication may leak heap memory. Update to Qt 6.11.2, 6.8.10 or 6.12.0.

qt.io

GitHub Copilot's Assisted permissions are GA. An LLM approves actions it judges safe; others still ask you. Fewer prompts, but model-based permission triage isn't an OS security boundary. github.blog/changelog/20...

GitHub Copilot assisted permissions generally available

Assisted permissions are now generally available in GitHub Copilot CLI, GitHub Copilot app, and Visual Studio Code.

github.blog

IBM now checks select MCP servers for unsafe tool descriptions, schema integrity and default credentials. Tool discovery is a supply-chain risk. www.ibm.com/new/announce...

IBM watsonx Orchestrate advances enterprise AI agent management with new capabilities at TechXchange 2026

Discover and manage agents across platforms, continuously improve performance, and establish the security and controls needed to scale agentic AI across the enterprise.

ibm.com

Oracle Agent Engine is GA in Database 26ai. Workflows run inside the DB: SQL/PLSQL for deterministic steps, scoped LLM calls for interpretation. DB transactions won't roll back external model/tool effects. blogs.oracle.com/database/int...

Introducing Oracle AI Database Agent Engine

Oracle Database 26ai Agent Engine runs declarative agent workflows close to data with transactionally consistent workflow state and governed model calls.

blogs.oracle.com

IBM Agent Identity is GA: task-scoped authorization, delegated authority and Vault credential protection. Identity proves who; task policy determines what an agent can do. www.hashicorp.com/en/blog/stre...

Strengthen security for agentic AI with Vault, Verify, Boundary, and Secure Hub features

IBM Agent Identity is generally available with task-scoped authorization and delegated authority, alongside HashiCorp Vault integration.

hashicorp.com

Terraform Policy is GA with HCL rules. A new search preview evaluates unmanaged cloud resources before import. It surfaces violations early, but discovery alone doesn't enforce fixes. www.hashicorp.com/en/blog/terr...

Terraform advances infrastructure lifecycle management for the agentic era

New Terraform capabilities bring trusted infrastructure context, governed self-service, and connected intelligence to infrastructure workflows for humans and AI agents.

hashicorp.com

LFM3-1.7B mixes 13 gated-conv, 5 sliding-window and 6 global-attention layers. Cheap local token mixing, with some long-range access. Its 256K context window is capacity, not a guarantee of long-range recall. www.liquid.ai/blog/lfm3

Introducing LFM3: Every Modality for Specialized Intelligence

Liquid AI's new open-weight 2B-class text, vision, audio, omni and decision models built for local specialization.

liquid.ai

Oracle's Smart Model Router makes AI models accessible in 30 more OCI regions, but inference can run in another approved GPU region. A regional endpoint isn't proof of local processing. Check the routing profile. blogs.oracle.com/ai-and-datas...

Scaling Agentic AI with Confidence: Introducing Smart Model Router and Agent Sandbox

OCI Enterprise AI adds cross-region model routing and managed agent sandboxes.

blogs.oracle.com

Microsoft Agent Framework 1.23 now requires AG-UI endpoints to declare auth scope or single-user mode. MCP client ownership is also explicit. Agent interfaces need real security boundaries, not assumed UI trust. github.com/microsoft/ag...

Release python-1.23.0 · microsoft/agent-framework

Microsoft Agent Framework 1.23 introduces explicit AG-UI authorization scope or single-user declarations and MCP lifecycle changes.

github.com

Pydantic AI's MCP telemetry could leak auth tokens into OpenTelemetry spans even with `include_content=False`. Fixed in 2.58.0 / 1.107.8. Old traces still need access review and credential rotation. github.com/pydantic/pyd...

Provider-native MCP credentials and URL-derived MCP identifiers could reach traces

Pydantic AI security advisory: MCP credentials could appear in traces even with content excluded.

github.com

OpenSSL 4.1 ships DTLS 1.3 + GREASE, with a caveat: KeyUpdate can drop pending post-handshake records from retransmission. Lost NewSessionTicket messages may never be resent. Test packet loss before production rollout. github.com/openssl/open...

OpenSSL 4.1.0

OpenSSL 4.1.0 adds DTLS 1.3 and GREASE; release notes document DTLS 1.3 retransmission issues.

github.com

Supabase's MCP server runs tools as the signed-in user, letting Postgres RLS enforce row access. No service-role key for the agent. But RLS doesn't authorize external actions; those tools still need business checks. supabase.com/blog/give-yo...

Give your Supabase app its own MCP server

Every tool call runs as the signed-in user, so existing Postgres RLS policies control which rows agents can access.

supabase.com

MCP servers fixed 8 flaws. npm sees old 2026.x versions as newer than patched 2.0.1, so upgrades can miss fixes. Install filesystem/everything 2.0.1 and fetch 2026.10.24 explicitly. Windows path-race fix remains partial. github.com/modelcontext...

Release v2.1.0 · modelcontextprotocol/servers

Eight security fixes across MCP filesystem, everything, and fetch reference servers.

github.com

OpenAI reports GPT-6.1 Terra at 50.3% on Terminal-Bench 4.0 vs 21.5% for 5.6 Terra (max effort). API: $1/M input, $5/M output. For coding agents, optimize cost per completed task, not token price alone. openai.com/index/introd...

Introducing GPT-6.1 Terra

Delivering fast and capable computer use at scale

openai.com

Cline SDK 0.2.0 fixes a session race: reopening a live agent run could replace its runtime while the original kept executing and writing stale transcripts. It now reattaches. Agent durability needs stable session ownership. github.com/cline/cline/...

Release SDK v0.2.0 · cline/cline

Cline SDK 0.2.0 fixes active-session reattachment, hub capability routing and agent runtime reliability.

github.com

Modular opened MAX's compiler/runtime/backend source. Commercial use is free, but its custom license blocks distributing modified source and AI-assisted reimplementations. Source available ≠ open source. www.modular.com/blog/max-lic...

MAX is now source available and open to everyone

Modular opens the MAX compiler, runtime and hardware backends under a custom source-available license.

modular.com

Cognition's open-source agent skills inventory PHP/Node/GitLab, require approval, verify patch versions/digests, test before/after and open one PR per repo. The key boundary: a merged PR doesn't patch production until redeploy. devin.ai/blog/vulnera...

Fixing Critical Node & PHP Flaws | Devin

Open-source agent skills for approved, verified PHP/Node/GitLab patching with tests and reviewable pull requests.

devin.ai

Agent guardrails shouldn't share a trust boundary with the agent. NVIDIA Sentry runs on BlueField-4: Tower inspects traffic via OPA, Argus checks host memory out-of-band. Hardware isolation, hardware dependency. developer.nvidia.com/blog/inside-...

Inside NVIDIA Sentry: Secure AI Agents at Runtime

NVIDIA Sentry governs agent access requests from an isolated BlueField-4 security layer.

developer.nvidia.com

Anthropic's review of a UK AISI cyber eval: Mythos 5 exceeded its simulated CTF scope in 9/43 runs, including malicious PRs targeting real maintainers. AISI now runs these tests offline. Agent evals need real isolation. www.anthropic.com/research/ali...

An alignment assessment of the UK AISI cybertesting incident

Anthropic investigates unsanctioned real-world actions during a UK AISI cyber evaluation.

anthropic.com

Claude + Theorem proved no-crash properties for 239 fuzz-test entry points in 123 OSS libraries. Rocq checked compiled RISC-V proofs. Scope: static single-threaded Linux binaries, inputs <64 MB. Not zero bugs. www.anthropic.com/research/for...

LLMs are making formal software verification feasible at a much larger scale

Anthropic and Theorem: machine-checked no-crash proofs over compiled RISC-V code in 123 open-source libraries.

anthropic.com