Tag

#security

Tagged “security”

18 articles

Pillar research says the AI coding agent sandbox leaks through trusted files

Pillar Security researchers demonstrated multiple sandbox-bypass techniques against AI coding agents, plus prompt-injection attacks hidden in READMEs, code comments and dependencies. OpenAI, Google and Cursor have patched several of the reported flaws.

Jul 23, 2026 · Maya Okonkwo Security & supply chain

Time-to-exploit went negative. Mandiant maps where AI helps the vulnerability queue, and where it fails silently.

Google Cloud's Mandiant team pins mean time-to-exploit at negative seven days and publishes a blueprint for aiming large language models at the vulnerability pipeline. The failure modes it lists are the ones that let the exploit through.

Jul 22, 2026 · Tomás Vega Security & supply chain

Secret scanning custom patterns get a REST API, so you can PR your detectors

GitHub made REST endpoints for secret scanning custom patterns generally available on July 13, 2026. Detectors can now be created, listed, updated and deleted via API at repository, organization and enterprise scope, though dry runs and the final publish step still happen in the UI.

Jul 21, 2026 · Priya Nair Security & supply chain

xAI publishes Grok Build's source after the coding agent was caught siphoning SSH keys

Three days after a researcher demonstrated the Grok Build CLI silently uploading SSH keys and repository contents to xAI's cloud, xAI dropped the agent's full source on GitHub under Apache 2.0. The code is now inspectable. The trust model is a different question.

Jul 21, 2026 · Tomás Vega Security & supply chain

The workstation is in scope now

A DevOps.com essay argues the laptop that pushes commits belongs inside the supply-chain threat model, citing the Megalodon campaign that injected malicious GitHub Actions workflows into thousands of repositories and a separate malicious VS Code extension incident. What CI-side controls actually catch this, and where the pipeline runs out of leverage.

Jul 19, 2026 · Tomás Vega Security & supply chain

OpenAI ships GPT-Red to automate prompt-injection testing against AI agents

OpenAI has released GPT-Red, a tool that automates prompt-injection testing against AI agents, per The New Stack. The framing is straightforward: once agents take real actions on real systems, manual red-teaming stops being enough.

Jul 18, 2026 · Tomás Vega Supply chain security

GitLab tries to auto-fix the transitive-dep problem it keeps quantifying

GitLab has moved Dependency Scanning Auto-Remediation into beta, aimed squarely at the transitive supply-chain risk it just spent a research cycle measuring. The interesting question is not whether the fix ships, but whether anyone downstream trusts the bot enough to merge it.

Jul 17, 2026 · Tomás Vega Code quality & testing

GitLab Duo Security Review Flow goes after the logic bugs SAST leaves on the floor

GitLab moved Security Review Flow into public beta on the Duo Agent Platform. You add it to a merge request like a human reviewer, and it focuses on authorization and business-logic bugs that pattern-based scanners typically miss.

Jul 17, 2026 · Priya Nair Developer experience

Autofix that reads more than one file: GitHub previews agentic code scanning fixes

GitHub's agentic autofix for code scanning alerts is in public preview, expanding remediation from the alert location out to the files around it. For CI/CD teams, a single alert now maps to a codebase-shaped change instead of a one-line patch.

Jul 14, 2026 · Priya Nair Security & supply chain

Datadog says months-long GitHub recon out of dormant 'ghost' accounts is prep for supply-chain attacks

Datadog security researchers say overlapping months-long campaigns have been abusing GitHub's public API and reactivating more than 50 long-dormant accounts to map organizations and their engineers. The setup looks a lot like the reconnaissance phase of a supply-chain intrusion.

Jul 14, 2026 · Maya Okonkwo Supply-chain security

The five-minute SBOM sniff test earns its keep

A quick, deliberate look at the bill of materials for a hardened image catches supply-chain trouble a CVE scanner will never flag. Bolt it into the pipeline, not the compliance report.

Jul 13, 2026 · Tomás Vega Security & supply chain

GhostApproval: six AI coding agents shipped a permission dialog that was reading the wrong path

Researchers at Google-owned Wiz disclosed GhostApproval, a class of flaw across six AI coding tools in which symbolic links redirect the agent's file writes while the human approval prompt shows the pre-resolution path. The mechanism is thirty years old; the CI/CD lesson is newer.

Jul 10, 2026 · Tomás Vega Supply chain security

npm v12 flips install-time to closed-by-default

npm v12 shipped as latest on July 8, 2026, with lifecycle scripts, git-URL dependencies and remote-URL dependencies all off by default at install time. Alongside it, the 2FA-bypass Granular Access Token starts a two-step retirement that ends direct publishes.

Jul 9, 2026 · Tomás Vega Supply-chain security

PolinRider keeps expanding, and the postinstall still lands on your runner

DevOps.com reports two DPRK-linked groups are expanding PolinRider, a supply-chain campaign that pushes malicious packages into developer workflows through long-running fake-interview scams. Socket and Rescana are named as the vendors doing the attribution. The pipeline lesson is old, and unfinished.

Jul 8, 2026 · Tomás Vega Supply-chain security

Aikido buys Root to patch open source in place, without the upgrade dance

Aikido Security acquired Root, a company whose technology fixes known vulnerabilities directly inside the package version you already run. The trade-off: someone other than upstream now decides what your dependency looks like on disk.

Jul 1, 2026 · Tomás Vega Supply chain security

npm freezes high-impact maintainer accounts for 72 hours after a sensitive change

npm now puts its 'high-impact' maintainer accounts into a 72-hour read-only state whenever it detects a sensitive account change like an email update or 2FA recovery code use. Publishing, tokens and team membership are frozen for the window; install and browse stay open.

Jun 26, 2026 · Tomás Vega Security & supply chain

Tenet's 'Agentjacking' research turns Sentry error data into a prompt-injection vector for AI coding agents

Security firm Tenet has demonstrated an indirect prompt-injection technique it calls Agentjacking, malicious instructions hidden inside Sentry error data that get executed by an AI coding agent with shell access. The operational read is that any third-party telemetry an agent ingests is now untrusted input.

Jun 16, 2026 · Maya Okonkwo Security & supply chain

SBOM attestation is becoming a default pipeline step

Generating and signing a software bill of materials at build time is shifting from compliance nice-to-have to standard CD hygiene. Here's the minimal viable setup.

Jun 12, 2026 · Tomás Vega