Security & privacy
Copilot Assisted permissions can auto-approve tool requests
Teams deciding how much command and file approval to automate.
NO FEED / Topics
Security & privacy
Permissions, credentials, dependencies and the handling of private code and data.
47 results
Security & privacy
Claude Code 2.1.309 repairs several sandbox and permission boundaries
Claude Code users relying on sandboxing, deny rules, MCP policy or unattended sessions.
Security & privacy
GitHub CLI can rotate OAuth tokens now. Audit anything that caches them first.
Pilot expiring credentials after removing every cached-token dependency.
Security & privacy
Cua Sandbox 0.9.1 rejects the retired cloud target
Computer-use agents and benchmarks still targeting Cua Cloud.
Security & privacy
Give a coding agent credentials without handing over production
Scope coding-agent access by task, resource and lifetime. Check inherited credentials, GitHub App tokens, OIDC trust and the actual sandbox boundary.
5 min readSecurity & privacy
Review an MCP server before connecting your coding assistant
Review an MCP server’s publisher, tools, permissions, data flow and update behavior. Use a disposable trial and verify how to revoke the connection.
6 min readSecurity & privacy
Run coding agents in GitHub Actions without trusting every issue
Review coding-agent workflows in GitHub Actions: untrusted issues and PRs, token permissions, runners, artifacts and the boundary before a privileged job.
6 min readSecurity & privacy
Verify AI-suggested packages before installing them
Check AI-suggested dependencies before installation. Separate package identity, provenance, hashes and script controls with practical npm and pip guidance.
6 min readSecurity & privacy
What leaves your repository when you use an AI coding assistant?
Compare specific Claude Code, Cursor and Copilot data paths: code transmission, indexing, training, retention, cloud sessions and feedback exceptions.
6 min readSecurity & privacy
Claude Code 2.1.291 fixes lost permission answers. Upgrade before diagnosing the user.
Move affected sessions to 2.1.291 and verify approval delivery plus quit/resume behavior.
Security & privacy
Claude Code 2.1.292 fixes sandbox read-boundary cases
Managed Claude Code environments and owners of subagent configurations.
Security & privacy
Claude Code 2.1.299 blocks a sandbox write path into reloadable plugins.
Upgrade, then check the plugin write boundary and the headless stop condition.
Security & privacy
Claude Code adds an explicit fail-closed hook option
Owners of policy hooks and unattended sessions with remote MCP servers.
Security & privacy
Claude Code mods can change the tool path. Review them like executable code.
Try an observable, bounded mod; test every behavior-changing hook before team rollout.
Security & privacy
Codex 0.162.0 repairs Linux sandbox boundary cases
Codex CLI users relying on Linux filesystem restrictions.
Security & privacy
Codex 0.163.0 restores policy on resume
Codex users with resumed tasks or managed network rules.
Security & privacy
Orca 1.4.221 fixes Copilot credential-file permissions
Orca users running GitHub Copilot on shared machines or SSH hosts.
Security & privacy
Pi 1.0 changes terminal defaults and MCP credential ownership
Pi users with terminal preferences or multiple MCP identities.
Security & privacy
VS Code 1.142 changes where managed agent permissions are enforced.
Test managed-policy routing before rolling 1.142 across a team.
Security & privacy
pydantic-ai 2.52.0 and 1.107.7 patch a denial-of-service in `web_fetch`
Anyone whose pydantic-ai agents fetch pages they do not control. Upgrade either line today.
Security & privacy
MCP TypeScript SDK 2.2.0 and 1.31.0 bind OAuth credentials to their issuer
Anyone running MCP clients against servers with long tool lists, or keeping OAuth tokens in a strict schema.
Security & privacy
GitHub Copilot Business and Enterprise admins have until 22 October to choose what new Copilot features do by default.
Pick a default this week, and set MCP servers and Code Review explicitly so neither depends on the global setting.
Security & privacy
The GitHub Copilot app can sandbox local sessions per project
Teams letting an agent run shell commands against a working tree that also holds credentials.
Security & privacy
GitHub is removing SHA-1 RSA SSH signatures and a Diffie-Hellman key exchange, on a dated schedule
Anyone whose CI, deploy keys or agent sandboxes speak Git over SSH from a pinned or old client image.
Security & privacy
Cline CLI 3.0.63 closes a Windows binary-hijack path and stops subagents re-asking for approval
Anyone running the Cline CLI on Windows, and anyone who has delegated work to configured subagents.
Security & privacy
pydantic-ai 2.44.0 and 1.107.6 patch four advisories
Anyone running pydantic-ai agents with webfetch enabled, or shipping OTel spans off-box.
Security & privacy
GitHub has disabled SHA-1 in HTTPS
Anyone whose pipeline still builds on an old base image. A TLS stack that only offers SHA-1 signature algorithms can no longer reach github.com, and it surfaces as a handshake failure in a clone that worked on Monday. Audit the oldest image in your CI first.
Security & privacy
Claude Code v2.1.271 · per-command allowed_domains, plugin commands pinned by hash
Anyone running Claude Code under an organisation policy, or installing plugins from a script
Security & privacy
Cline's hub loads agent plugins from your home directory and deliberately refuses to load them from the repository you just opened.
Upgrade. On a large workspace, or on a provider that forwards 429s, this removes two failure modes you may have been blaming on your own machine.
Security & privacy
Copilot · enterprise managed agent permissions (GA)
GitHub Copilot Business/Enterprise admins rolling out agent workflows
Security & privacy
MCP Python SDK 2.2.0 · origin redirects, 30 min idle, OAuth issuer
Anyone hosting or calling MCP over HTTP with the official Python SDK
Security & privacy
MCP Python SDK 2.2.0 tightens redirects, idles sessions at 30 min, and checks OAuth issuer.
Read 2.2.0 before upgrading MCP Python 2.x — redirects, idle sessions, and OAuth issuer checks change defaults. 1.x operators: ≥1.30.0. MAI-Code-1-Flash leaves Copilot 10 Sep.
Security & privacy
Claude Code 2.1.259 · managed MCP + headless
Org Claude Code fleets and unattended hosts
Security & privacy
Claude Code 2.1.248 · --restricted, hourly cache fix, cross-session messaging
anyone running Claude Code in CI, shared machines, or multi-hour sessions
Security & privacy
Claude Code 2.1.248 adds --restricted; npm latest is 2.1.250 while stable stays 2.1.236.
latest ≥2.1.248 for --restricted and the hourly cache fix; pin stable only if you need the slower channel.
Security & privacy
The session left the laptop
Closing a laptop can stop access to its folders while a cloud task keeps running. The August releases moved work across that boundary; the permissions need to make the same journey.
6 min readSecurity & privacy
openai-agents-python 0.22.0 · guardrail redaction + explicit-client contract
OpenAI Agents SDK users with guardrails, multi-checkpoint runs, or explicit client wiring
Security & privacy
Claude Code 2.1.234 · auto-continue at usage-limit reset + NTLM path harden
Claude Code daily drivers on Pro/Max/Team and multi-session hosts
Security & privacy
openai-agents-python 0.21.1 · model call timeouts + sandbox controls
OpenAI Agents SDK users who need hard call budgets or tighter sandboxes
Security & privacy
The click was never the control
August’s switch to automatic command screening addressed approval fatigue. A tax-calculation fix shows why the place where code reaches production still needs its own decision.
5 min readSecurity & privacy
Claude Code auto mode is the default for new Pro, Max, and Team sessions as of today.
If you never set a default, decide before the next session: keep auto and pin ask rules on git push / gh pr create, or set "defaultMode": "manual" in ~/.claude/settings.json. Org admins who want the old baseline set permissions.disableAutoMode to disable.
Security & privacy
pydantic-ai 2.30.0 / 1.107.5 · Host-header fix on local web UI
anyone who runs the pydantic-ai dev web chat on a workstation
Security & privacy
Claude Code 2.1.229 · auto mode fix + marketplace command sources
auto-mode users without attribution headers; plugin authors
Security & privacy
pydantic-ai 2.28.0 · web UI CSRF fix
anyone who runs the pydantic-ai dev web chat on a workstation
Security & privacy
Vercel Sandbox · Managed Images
anyone spinning agent sandboxes on Vercel who was baking tools at boot
Security & privacy
Claude Code makes auto mode the default on 14 Aug. The number that sold it: humans caught 13.6% of planted dangerous commands.
already on auto mode — only the free classifier is new. On manual or YOLO, decide before the 14th: Shift+Tab still flips modes; admins can pin or disable auto. Do not read 89% as solved; eleven percent still gets through.
Security & privacy
A cyber-range agent opened a malware PR on a real open-source project and social-engineered the maintainer.
do not treat this as a product recall. Treat it as a reason to keep approval gates and network egress on any agent that can open PRs or message people, including the ones you already trust.