GitHub (Organization): news timeline & CVEs

New Aoki Infostealer Targets AI Coding Assistant Credentials

Symantec says the Aoki infostealer now steals API keys and sessions from AI coding assistants.

Symantec's Threat Hunter Team says newer builds of the Aoki infostealer steal credentials and session files from AI coding assistants, expanding beyond browser, cryptocurrency-wallet, and password-manager theft. Targeted files include .claude.json, Claude desktop cookies, ChatGPT cookies, and Codex config.toml and .env files, plus tokens for GitHub Copilot, Cursor, Gemini, JetBrains AI Assistant, and Kiro, and AWS credentials. Stolen keys can support LLMjacking and expose linked repositories, cloud services, and CI/CD pipelines. Aoki was first observed in 2026; Symantec has not tied it to a known group and assesses the operators as financially motivated.

Infosecurity Magazineupdated 10:22 UTC · 7h agofirst Oct 15, 13:00 UTC · 1d agoMalware in the wild 3 sources3

10 Best Secrets Detection Tools Compared (2026): Features & Pricing

A 2026 comparison ranks ten secrets-detection tools, from GitGuardian and TruffleHog to free GitHub and Gitleaks scanning.

GBHackers compares ten secrets-detection products, separating dedicated platforms, verification engines, native controls, and cloud or SaaS scanners. GitGuardian leads the platform lane, TruffleHog is cited for live-secret verification, and GitHub Secret Scanning plus Gitleaks are presented as the free floor. Nightfall, Wiz, Snyk, Aikido, Check Point Spectral, and 1Password cover collaboration, cloud context, and vault-adjacent workflows. The comparison is research-based and names no new vulnerability.

GBHackersOct 15, 07:04 UTC · 1d agoTools 2 sources1· 1 read

Top 10 Best DevSecOps Platforms in 2026 [Ranked & Scored]

A 2026 ranking puts GitLab, GitHub Advanced Security, and Snyk atop DevSecOps platforms by developer fix rates.

A 2026 editorial ranking scores ten DevSecOps platforms mainly on whether developers remediate findings, not scanner volume. GitLab leads at 9.1, followed by GitHub Advanced Security at 9.0 and Snyk at 8.9. Aqua, JFrog, Prisma Cloud, Checkmarx, Harness, Datadog, and Cycode round out the list. The methodology cites no lab testing and weights adoption, coverage, pipeline nativity, pricing clarity, and runtime connection.

Cyber Security NewsOct 15, 07:02 UTC · 1d agoIndustry1

AI-generated changes create review bottlenecks in software delivery

JetBrains finds most DevOps teams run CI/CD, while AI-generated changes slow code reviews.

JetBrains’ State of CI/CD in 2026 report says 83.4% of respondents who use DevOps practices run CI/CD pipelines, ahead of containerization, infrastructure as code, and GitOps. GitHub Actions leads organizational adoption, followed by GitLab CI and Jenkins, and 72.8% self-host their primary system. About 48.8% use AI for at least one CI/CD-related task, but only 8% run AI-powered steps inside a build or test, and teams report that generated changes create review bottlenecks.

Help Net SecurityOct 15, 04:30 UTC · 1d agoIndustry

ASPM in the AI Era Part 1: Unified Visibility Across Your Stack

Wiz promotes ASPM connectors that centralize scanner findings and prioritize exploitable application risks.

Wiz's first post in an ASPM series argues that AI-assisted development and faster exploit generation make manual AppSec triage inadequate. It describes ingesting SAST, SCA, DAST, and secrets findings through Wiz Code connectors for tools such as Snyk, Checkmarx, and GitHub Advanced Security, plus SARIF uploads. Using React2Shell (CVE-2025-55182) in Next.js as an example, Wiz says context should distinguish internet-facing or sensitive instances from thousands of duplicate package alerts. The post focuses on product setup and asset correlation rather than a new incident.

Wiz BlogOct 14, 20:13 UTC · 1d agoIndustry CVE-2025-551822· 1 read

GitHub’s model for secret detection: Finding more secrets with less noise

GitHub's ModernBERT secret classifier raised alert acceptance 11.5% and cut median scan time by 86%.

GitHub described a two-stage secret scanner that uses regex candidate generation and a fine-tuned ModernBERT classifier to separate real credentials from lookalikes. Compared with its previous LLM pipeline, GitHub reports an 11.5% rise in owner-weighted alert acceptance, an 86% reduction in median post-push scan time, and more than 98% lower serving cost. The efficiency gain let GitHub extend generic secret detection to push protection so findings can be returned before a push is accepted.

GitHub Blog · Securityupdated Oct 15, 07:04 UTC · 1d agofirst Oct 13, 17:56 UTC · 2d agoTools 2 sources1· 1 read

GitHub Tightens npm Publishing After Flaw Left 1,500+ Packages Vulnerable to Hijacking

GitHub expires unused npm trusted-publisher configs after 48 hours following a slug-reuse flaw affecting 1,500-plus packages.

Bloom Security told GitHub on June 6, 2026, that npm trusted publishing anchored trust to a repository slug rather than GitHub's immutable repository ID until the first successful publish. An attacker who reclaimed a freed slug could exchange a GitHub Actions OIDC token for a short-lived credential and publish a malicious version of a package they did not own, an issue dubbed SlugFest. Researchers found more than 1,500 exposed packages with over 10 million collective downloads. GitHub now expires unvalidated trusted-publisher configurations after 48 hours and rejects tokens from issue_comment events.

The Hacker NewsOct 13, 16:18 UTC · 3d agoVulnerability 2 sources1· 1 read

All Systems Go: Global PC App Ecosystem Ready for NVIDIA RTX Spark

NVIDIA prepares RTX Spark PCs for an Oct. 16 launch with local AI agents and Nemotron 3.5 Bolt 74B.

NVIDIA says RTX Spark PCs, combining gaming, creative, and local AI compute, will be available Oct. 16 after a launch announcement with Microsoft. The platform has up to 128GB of unified memory and 1 petaflop of AI compute, with CUDA, PyTorch, and agent apps such as GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi, and WorkBuddy being optimized. Nemotron 3.5 Bolt 74B-A9B, a 74B mixture-of-experts model with 9B active parameters quantized to 24GB, is due Oct. 15 for local agents. Adobe apps, Broadcast 2.3, ComfyUI, Blender, and DLSS updates are also planned.

NVIDIA Blogupdated 15:00 UTC · 2h agofirst Oct 13, 12:00 UTC · 3d agoAI industry 5 sources1· 1 read

Over 1,500 npm packages exposed by Trusted Publishing validation flaw

A now-fixed npm Trusted Publishing flaw exposed over 1,500 packages to repository-name hijacks.

Bloom Security disclosed SlugFest, a flaw in npm Trusted Publishing that verified publishers by human-readable GitHub repository names instead of permanent repository IDs. An attacker who reclaimed an abandoned name could obtain an OIDC publishing credential and release a package version without compromising the maintainer. Researchers found more than 1,500 potentially exposed packages with over 10 million combined downloads. npm shipped a fix on October 2 that expires unvalidated configurations after 48 hours; Bloom reported no in-the-wild exploitation.

CSO Onlineupdated Oct 13, 16:18 UTC · 3d agofirst Oct 13, 10:02 UTC · 3d agoVulnerability 2 sources1

Incident with Actions, Pull Requests and Issues

GitHub mitigated degraded Actions, Issues, and Pull Requests, with notification delays still being fixed.

GitHub's status page reported degraded performance on October 12, 2026 for Actions, Issues, and Pull Requests. Users had trouble loading the Actions tab, using the Checks API, queueing workflow runs, creating or updating issues, and requesting pull-request reviews through REST and GraphQL. Pull Requests recovered first, and later updates said Actions, Issues, and Projects were mitigated while notification delays were still being addressed. The incident report does not describe a security compromise.

Hacker News · AIOct 12, 12:23 UTC · 4d agoOther HN 45↑ · 30 comments2

Python 3.15.0 added to actions/python-versions

Python 3.15.0 is now available in GitHub Actions python-versions for CI test matrices.

GitHub's actions/python-versions repository now includes the stable Python 3.15.0 release. Developers can set "3.15" in a GitHub Actions testing matrix to run CI against the new interpreter. Simon Willison said he had been waiting for the addition and had asked ChatGPT to check the repository hourly until it landed.

Simon WillisonOct 10, 23:58 UTC · 5d agoOther1· 1 read

GhostAction Hackers Compromise 500+ GitHub Accounts to Steal Cloud and AI API Credentials

GhostAction compromised 500-plus GitHub accounts and planted workflows that steal cloud and AI API credentials.

Socket reports the GhostAction campaign has compromised more than 500 GitHub accounts and injected malicious GitHub Actions workflows into tens of thousands of repositories since October 7, 2026. Workflows such as security-audit.yml run on push, steal Actions secrets, and scan git history for AWS, Anthropic, OpenAI, OpenRouter, GitHub, GitLab, Google, Slack, and SendGrid credentials. Stolen data is sent by cleartext HTTP POST to 193.32.204.199. Successful runs were confirmed, including against kitao/pyxel and uber/athenadriver, but no malicious PyPI or crates.io releases have been observed.

GBHackersupdated Oct 15, 08:45 UTC · 1d agofirst Oct 10, 09:02 UTC · 6d agoThreat actor in the wild 3 sources6

New GhostAction Attack Compromises Hundreds of GitHub Repos to Steal Secrets

GhostAction hit 346 GitHub repositories, including Uber's, with malicious Actions workflows that stole CI secrets.

Socket said the GhostAction campaign compromised 346 GitHub repositories on October 8 after hijacked maintainer accounts henrywoo and kitao added a fake security-audit GitHub Actions workflow. The workflow posts CI/CD secrets to 193.32.204[.]199 and scans files and full Git history for cloud keys, tokens, and API credentials, including AWS, GitHub, GitLab, Google, Slack, and major AI-provider keys. Targets include Uber's uber/athenadriver and kitao/pyxel, which has more than 18,000 stars. GitGuardian previously tied the campaign to 817 repositories in September 2025 and 772 more through September 2026; Socket had not seen poisoned PyPI or crates.io releases from this wave as of October 9.

Cyber Security NewsOct 9, 16:38 UTC · 7d agoThreat actor in the wild5· 4 reads

Vulnerable Models in the AI Supply Chain: An Empirical Study of Adoption, Propagation, and Usage Patterns

A study finds 3,757 flagged vulnerable Hugging Face models referenced across more than 5,000 repositories without sanitization.

Researchers tracked 16,977 Hugging Face models flagged by InsightDB for issues including insecure serialization and architectural backdoors. Despite public flags, those models appear in more than 21,000 code locations across more than 5,000 open-source repositories, and 3,757 distinct models (22.1%) are referenced by at least one project. Using code property graphs and frequent subgraph mining, they found recurring integration patterns but no provenance verification or sanitization. Developers often bypassed security parameters or trusted permissive framework defaults.

arXiv cs.CROct 9, 02:27 UTC · 7d agoAI safety & security1· 1 read

FakeGit malware campaign returns with 17,610 malicious GitHub repos

FakeGit returned with 17,610 GitHub repositories distributing SmartLoader and the StealC infostealer.

Apiiro says the FakeGit campaign resumed on October 4 and is using 17,610 GitHub repositories to distribute SmartLoader, which then delivers the StealC infostealer. In 34 hours the operator pushed more than 13,000 repositories, peaking at 2,999 an hour. In sampled commits, 97% changed only the README and 88% pointed a Download button at a ZIP that installs SmartLoader. Island previously tied the operation to 7,600 fake repositories in July, including about 800 posing as AI skills or MCP servers.

BleepingComputerOct 8, 17:10 UTC · 8d agoMalware in the wild1

How one bug bounty researcher chooses the features they investigate

GitHub profiles VIP bug bounty researcher @vaib25vicky and higher payouts for quality findings.

For Cybersecurity Awareness Month, GitHub's bug bounty team interviewed VIP researcher @vaib25vicky, who specializes in authorization and access-control issues. GitHub says its restructured program pays more for higher-quality reports, with VIP payouts of up to $30,000 or more for critical findings and faster response times. Researchers may be invited after one critical, two high, four medium, or seven low resolved findings. The researcher says most high-impact bugs remain authorization issues and that AI output must be verified before submission.

GitHub Blog · SecurityOct 8, 17:02 UTC · 8d agoIndustry2· 1 read

Tensorlake npm Package Compromised to Spread Shai-Hulud Worm and Steal Developer Secrets

Malicious tensorlake@0.5.144 on npm delivers a Shai-Hulud worm that steals developer secrets and can wipe hosts.

Aikido reported that npm package tensorlake version 0.5.144, published on 8 October 2026, contains a new Shai-Hulud worm variant. A preinstall script downloads the Bun runtime and runs an obfuscated payload that harvests cloud, CI/CD, SSH, Vault, GitHub, browser, and cryptocurrency-wallet secrets. Stolen data can be sent to iseekaigogo[.]com or to an address retrieved from an actor-controlled Ethereum contract. The payload can also wipe infected machines if an embedded GitHub token is revoked; PyPI and Cargo distributions were not compromised.

Cyber Security NewsOct 8, 15:50 UTC · 8d agoMalware in the wild7· 5 reads

GitHub AI Can Spot Passwords Hidden in Code Before Developers Push Them

GitHub is previewing a ModernBERT classifier that blocks hidden passwords during pushes, beyond pattern-based scanning.

GitHub and Microsoft Applied Sciences introduced a ModernBERT classifier that uses surrounding code to detect likely passwords before a push, extending protection beyond known token patterns. GitHub says candidate batches are checked in under two milliseconds and that the model could more than double secrets blocked. From Q2 2024 through Q2 2026, screened public pushes rose 2.84 times, the share containing secrets showed no clear trend, and push-protection overrides fell from 6.63% to 3.93%. The capability is in private preview for Secret Protection customers on Enterprise Cloud and GitHub Team and will consume AI credits.

Cyber Security NewsOct 8, 14:01 UTC · 8d agoAI tools & infra2· 1 read

Hackers Hijack Tensorlake Package to Spread Shai-Hulud Supply Chain Malware

Attackers published a malicious tensorlake npm package that spreads a new Shai-Hulud supply-chain worm.

A threat actor published tensorlake npm version 0.5.144 on October 8, 2026, embedding a new self-replicating Shai-Hulud worm. Its preinstall hook runs obfuscated JavaScript that downloads the Bun runtime and steals environment variables, npm and GitHub tokens, cloud and SSH keys, Kubernetes and CI/CD secrets, and data from 14 cryptocurrency browser extensions. Stolen publishing credentials can spread the worm, and a dead-man’s switch can wipe hosts if an embedded GitHub token is revoked. The package was seeded via commits on Tensorlake’s GitHub; PyPI and Cargo builds were not compromised, and the npm package has over 100,000 lifetime installs.

GBHackersOct 8, 13:30 UTC · 8d agoMalware in the wild5· 1 read

GhostAction Supply Chain Campaign Uses Malicious GitHub Actions to Steal CI/CD Credentials

GhostAction hit 772 GitHub repositories with malicious Actions workflows that stole CI/CD secrets.

GitGuardian reported a GhostAction wave from August 31 to September 30, 2026, that planted malicious GitHub Actions workflows in 772 public repositories and targeted 2,577 secrets across 373 users and organizations. Stolen GitHub access was used to commit workflows that read referenced secrets and exfiltrated them over plain HTTP to 193.32.204.199. Reviewers found 336 successful runs that stole 26 secrets from 13 repositories, and only 16% of affected repositories were cleaned by October 5. A 2025 wave previously hit 817 repositories and at least 3,325 secrets.

Cyber Security NewsOct 8, 12:32 UTC · 8d agoThreat actor in the wild2· 1 read

Hackers Use GitHub-Hosted Poem to Control PoeLLM Malware Targeting AI Infrastructure

PoeLLM malware uses a GitHub poem to derive C2 addresses and mine cryptocurrency on more than 3,400 exposed AI servers.

Lumen’s Black Lotus Labs says PoeLLM, active since at least April 2026, has compromised more than 3,400 servers, mostly in the United States and Western Europe. The malware reads four marked words from a poem in a GitHub repository and maps them to the current command-and-control IPv4 address, so operators can rotate infrastructure by editing the poem. It targets exposed LiteLLM, Ollama, Gotenberg, and Gitea services; one LiteLLM path is linked to command-injection flaw CVE-2026-42271. The Linux payload provides a remote shell, scanning, exploit delivery, and XMRig and Iron miners that contact Kryptex.

Cyber Security NewsOct 8, 09:14 UTC · 8d agoMalware in the wild 6 sources CVE-2026-422715· 2 reads

GitHub adds AI to catch passwords before a code push

GitHub is adding a ModernBERT classifier to push protection to block unstructured passwords before they enter repositories.

GitHub and Microsoft Applied Sciences built a ModernBERT classifier that expands push protection beyond recognizable credential formats to unstructured secrets such as database passwords. GitHub says the model scores batches in under two milliseconds and could more than double the secrets push protection blocks. From Q2 2024 to Q2 2026, screened public pushes rose 2.84 times and credential-bearing pushes rose 2.59 times, while about 30 percent of newly detected secrets are blocked before they enter history. The capability is in private preview, with a later-October rollout for GitHub Secret Protection on Enterprise Cloud and Team plans, and it will ship in GitHub Enterprise Server 3.23.

Help Net SecurityOct 8, 09:02 UTC · 8d agoTools2

Top 10 Best Secrets Detection Tools in 2026 [Ranked & Scored]

A 2026 ranking of secrets-detection tools puts GitGuardian first, then TruffleHog and GitHub.

Cyber Security News ranked ten secrets-detection tools for 2026, weighting live verification highest in an editorial, non-lab scoring model. GitGuardian placed first, Truffle Security’s TruffleHog second, and GitHub secret scanning third. Wiz was scored for cloud blast-radius context and Nightfall AI for SaaS coverage beyond git, with free options such as Gitleaks hooks and TruffleHog OSS noted.

Cyber Security NewsOct 8, 07:34 UTC · 8d agoIndustry3· 1 read

12 Best SCA Tools Compared (2026): Features & Pricing

A 2026 SCA comparison highlights Snyk, Sonatype, and Socket across three dependency-risk lanes.

GBHackers compared twelve software composition analysis tools, arguing dependency risk spans known CVEs, malicious packages, and license exposure. Snyk leads the developer-platform lane, Sonatype the repository-firewall lane, and Socket malicious-package detection. Mend is highlighted for Renovate-based remediation, Endor Labs for function-level reachability, and FOSSA and Black Duck for license and legal depth.

GBHackersOct 8, 07:33 UTC · 8d agoIndustry 5 sources5· 3 reads

[AINews] Claude Haiku 5.5 — better than GPT-6 Luna at the same pricing

Anthropic released Claude Haiku 5.5, a cheaper small model priced to match GPT-6 Luna.

Anthropic released Claude Haiku 5.5, its first Haiku update in about a year, priced to match OpenAI’s GPT-6 Luna and described as roughly 75% cheaper to run than Haiku 4.5. Below 100K prompt tokens it costs $0.10 per million input and $0.50 output; above that threshold the rates are $0.50 and $2.50. Artificial Analysis reported an Intelligence Index of 43, ahead of GPT-6 Luna at 38, a 1 million-token context, and 33% on Terminal-Bench 4.0. Anthropic also halved Sonnet 5.5 cache-read prices and is fixing a pre-release bug that caused over-refusal.

Latent SpaceOct 8, 07:27 UTC · 8d agoModel release 6 sources6· 4 reads

Hackers Abuse GitHub Actions to Steal SSH Keys, Cloud Credentials and Access Tokens

GhostAction abused GitHub Actions workflows to steal SSH keys and cloud credentials from public repositories.

GitGuardian’s GhostAction investigation found 772 public repositories across 373 GitHub users and organizations compromised between August 31 and September 30, 2026. Attackers added workflows such as github_actions_security.yml that posted named secrets, including SSH private keys and cloud credentials, over unencrypted HTTP to 193.32.204.199. Researchers confirmed 26 secrets exfiltrated from 13 repositories while 2,577 were targeted, and only about 16% of affected repositories showed effective cleanup by October 5. The activity continues earlier GhostAction and Shai-Hulud campaigns; one related repository also contained an XMRig miner.

GBHackersOct 8, 06:11 UTC · 8d agoThreat actor in the wild6· 4 reads

Malicious npm Packages Steal Browser Passwords, Discord Tokens and Crypto Wallets.

The MALFEX npm campaign ships Overlord RAT and movinlike to steal browser passwords, Discord tokens, and crypto wallets.

Checkmarx describes MALFEX, an npm supply-chain campaign active since August 2023 that shipped Windows malware in eight packages totaling 40,767 downloads by October 1, 2026. Install scripts and package-load hooks deliver Overlord RAT, which uses a scheduled task named Maiden, and a Go downloader for the movinlike Node.js stealer. movinlike targets Discord clients, browser cookies and passwords, Telegram Desktop sessions, and wallets including MetaMask, Phantom, and Coinbase Wallet, then exfiltrates data through a Discord webhook. Three packages remained installable, and OSV advisories did not cover every malicious version.

GBHackersupdated Oct 8, 07:51 UTC · 8d agofirst Oct 8, 04:53 UTC · 8d agoMalware in the wild 4 sources6· 3 reads

Skill Constellations: Tracing the Supply Chain of Agent Skills on GitHub

Study of 2.2 million GitHub agent-skill copies finds stars miss the repositories that spread risk.

Researchers build the first dated copy network of agent skills from the git history of every SKILL.md in GitSkills, covering 2,193,119 adoptions on GitHub. Skills are instructions and scripts that agents such as Claude Code and Codex run with the user's permissions, and copies rarely pick up fixes made at the source. A model of which repositories others copy from shows that reviewing its top 100 prevents 14.9% of later high-risk skill adoptions, versus 0.5% for the 100 most starred. The authors recommend distributing versioned references instead of copies.

Hugging Face daily papersOct 7, 20:00 UTC · 8d agoAI safety & security2

GitHub Incident with Git Operations, Pull Requests and Actions

GitHub reported a short outage degrading Git operations, pull requests, Actions, webhooks, and issues.

GitHub’s status page reported degraded service beginning around 15:14 UTC on October 7, 2026. Widespread impact was observed from 15:06 to 15:16 UTC across Git Operations, Webhooks, Issues, Pull Requests, and Actions. Later updates said Actions and pull requests recovered and Git operations briefly returned to normal, while investigation of remaining degradation continued. No security cause was stated.

Hacker News · AIOct 7, 15:17 UTC · 9d agoOther HN 64↑ · 25 comments2· 1 read

PoeLLM malware has assembled a sweeping botnet, taking technical cues from a poem

PoeLLM malware has compromised over 3,400 servers, encoding its command-and-control address in a GitHub poem.

Lumen Technologies’ Black Lotus Labs said PoeLLM malware has compromised more than 3,400 servers since April by targeting exposed services including LiteLLM, Ollama, Gotenberg, and Gitea. The malware reads a poem on GitHub and maps four words through a hard-coded dictionary to assemble its command-and-control IP, so operators can rotate infrastructure without updating implants. The botnet conducts exploit scanning and cryptocurrency mining and can run remote code, which researchers say could abuse AI models on victim servers. Italian-language comments and Italy-based servers suggest an Italian-speaking operator, with no observed links to known groups.

CyberScoopupdated Oct 8, 09:14 UTC · 8d agofirst Oct 7, 15:00 UTC · 9d agoMalware in the wild 6 sources2

Inside the Advisory Database and what happens when vulnerability volume breaks records

GitHub says record vulnerability volume — 1,560 reviewed advisories published in May 2026 — has stretched advisory review times to weeks, urging better report quality.

GitHub's Advisory Database published 1,560 reviewed advisories in May 2026, more than five times typical monthly output, while private vulnerability reports grew from roughly 550/week in January to over 3,000/week. Repository advisories scaled from about 650/week to more than 5,000/week, GitHub CNA CVE requests reached almost 4,000 in May (nearly 10x year-over-year), and the CVE program has published over 30,000 CVEs in 2026. Since mid-April, publication times have stretched to multiple weeks for a meaningful share of advisories, which can widen exposure windows. GitHub asks reporters to submit complete vulnerability data, coordinate with maintainers, and request CVEs only with intent to publish.

Lobsters · securityOct 7, 13:39 UTC · 9d agoIndustry3· 1 read

Encrypted instructions trick Copilot CLI into spilling developer secrets

Adversa AI shows encrypted prompts can make GitHub Copilot CLI read local secrets and exfiltrate them.

Adversa AI described Cryptographic Context Injection, in which encrypted instructions fetched by GitHub Copilot CLI are decrypted inside the agent and then treated as trusted context. In a demonstration, Copilot read a .env.prod file and sent its contents to an attacker endpoint in 28 seconds without naming the destination. The chain requires autopilot mode and a model that follows the decrypted instructions; mai-code-1.1-flash completed it in half of runs, while two GPT-5.6 models refused. GitHub validated the report but said it is not a vulnerability or bounty-eligible because the user granted autonomous permissions and asked the agent to fetch attacker-controlled content.

CSO OnlineOct 7, 11:47 UTC · 9d agoAI safety & security 4 sources2· 1 read

AI Agent Gateway: Open-source tool keeps credentials out of agent configs

Tuskira's open-source AI Agent Gateway keeps credentials out of agent configs and enforces tool permissions.

Tuskira released AI Agent Gateway, a free open-source proxy between AI agents, MCP tool servers, and model providers. Agents send a gateway key and profile; the gateway authorizes the call, then attaches credentials from an encrypted store so the agent never holds tokens such as a GitHub secret. Model traffic for Anthropic, AWS Bedrock, OpenAI, and Gemini can use the same gateway, which records token use and estimated cost. Unbound keys can request any profile, and optional storage of prompt and response bodies, capped at 1 MiB, can be turned off.

Help Net SecurityOct 7, 05:30 UTC · 9d agoAI tools & infra1

A sustainable web career, for when all this blows over

A web developer urges accessibility, CSS, and communication skills over React, GitHub, and influencers.

David Bushell's essay advises front-end developers to focus on accessibility, CSS, and communication rather than React. He calls GitHub a liability and recommends self-hosted forges such as Forgejo, plus local or independent CI/CD. The piece is career opinion and also urges readers to reject far-right politics in tech.

Hacker News · securityOct 6, 19:31 UTC · 9d agoOther HN 39↑ · 44 comments2

Tell HN: GitHub refuses to remove cracked copies of my software after a month

A Hacker News poster says GitHub still hosts cracked copies of their software after a month.

A Tell HN post says GitHub has not removed cracked copies of the author's software after about a month of requests. The submission showed 25 points and 15 comments. The item text does not name the software, cite a takedown notice, or include a GitHub response.

Hacker News · securityOct 6, 18:54 UTC · 9d agoOther HN 25↑ · 15 comments

Hackers Pose as Dubai Airports Recruiters to Infect Software Engineers With ShelbyLoader V2

Iranian-linked CL-STA-1178 posed as Dubai Airports recruiters to infect engineers with ShelbyLoader V2.

Unit 42 says Iranian state-aligned actor CL-STA-1178 impersonated Dubai Airports and sent software engineers a Visual Studio coding assessment. Opening the weaponized project abused MSBuild, AppDomainManager hijacking, and DLL sideloading to run ShelbyLoader V2 inside a renamed, Microsoft-signed host. The loader beaconed through a GitHub repository, decrypted ShelbyC2 in memory, and a Blackwood loader reflectively ran Chisel for tunneling. Researchers reported no compromise of Dubai Airports, and GitHub removed the infrastructure; related Blinder Tunnel activity has targeted Iraqi critical infrastructure since March 2026.

GBHackersupdated Oct 6, 14:31 UTC · 10d agofirst Oct 6, 13:23 UTC · 10d agoThreat actor in the wild 3 sources3· 1 read

Meta and Microsoft Reduce Internal Use of Claude AI

Meta and Microsoft are cutting internal Claude use while pushing their own coding assistants.

The Information reported that Meta and Microsoft are steering employees away from Anthropic's Claude toward their own coding assistants. Microsoft's forecast internal Anthropic spend, once above $1 billion a year, reportedly fell by more than a third, and many cloud and AI employees saw monthly limits cut from $100,000 to about $10,000, while customer use of Anthropic models on Microsoft platforms still grew. Meta's Claude Code users reportedly fell from about 60,000 to 30,000 as MetaCode and Muse Code gained users, even after Meta spent over $105 million on Claude Code in 28 days. The report also cites previously patched Claude Code issues CVE-2025-59536 and CVE-2026-21852 and Microsoft's RoguePilot Codespaces flaw as reasons to govern agent permissions.

GBHackersupdated Oct 7, 18:49 UTC · 8d agofirst Oct 6, 12:26 UTC · 10d agoAI industry 3 sources CVE-2025-59536 CVE-2026-21852

Blinder Tunnel Campaign Targets Iraqi Infrastructure

Unit 42 attributes the Blinder Tunnel campaign to Iranian state-aligned CL-STA-1178, which impersonated Dubai Airports IT to hit Iraqi critical infrastructure.

Unit 42 tracks an Iranian state-aligned threat actor, CL-STA-1178, running the Blinder Tunnel campaign against Iraqi critical infrastructure from March 2026, with infrastructure staged since November 2025. The actor used fake Dubai Airports recruitment lures delivering trojanized coding challenges, then a three-step chain: weaponized .csproj files, AppDomainManager hijacking, and DLL sideloading to deploy ShelbyLoader V2 malware. C2 abused GitHub API repositories for decryption keys, payload delivery, and GitHub issues as fallback. OpSec errors linked the campaign to Google Drive credential harvesting against an Israeli entity in May-June 2026; GitHub has taken down the malicious infrastructure.

Palo Alto Unit 42updated Oct 6, 14:31 UTC · 10d agofirst Oct 6, 10:00 UTC · 10d agoThreat actor in the wild 3 sources

GitHub’s ReviewBench puts AI code reviewers to the test

GitHub launched ReviewBench, a 219-pull-request benchmark for testing AI code-review agents.

GitHub released ReviewBench in research preview to compare AI code-review agents on 219 pull requests from 187 public repositories and 19 programming languages. A three-stage golden set merges findings from human reviewers, follow-up commits, analysis tools, and models, then scores them with a shared rubric. Grounded precision, recall, and F1 measure overlap with known issues, while augmented metrics use an AI judge for additional findings. GitHub said a Copilot code-review lite ensemble raised production recall 13.6 percent, increased comments that prompted code changes by 8 percent, and cut cost per review 8 percent versus a single-reviewer control.

Help Net SecurityOct 6, 09:29 UTC · 10d agoAI research

Top 10 Best SAST Tools in 2026 [Ranked & Scored]

A 2026 editorial ranking puts Snyk Code first among ten SAST tools, ahead of CodeQL and Semgrep.

Cyber Security News published an editorial 2026 ranking of ten SAST tools, weighting fix-rate potential at 30% and precision at 25%, and said the scores were not lab-tested. Snyk Code leads at 9.1; the article names a podium of Snyk, GitHub CodeQL, and Semgrep, while its numbered table places unscored Qwiet AI second and Semgrep third at 8.8. Later scores include Checkmarx 8.7, SonarQube 8.6, Veracode 8.4, Black Duck Coverity 8.2, OpenText Fortify 8.0, and HCL AppScan 7.8, with DeepSource unscored.

Cyber Security NewsOct 6, 07:45 UTC · 10d agoIndustry6· 4 reads

GitHub

Timeline114 articles

Related CVEsby mentions

Appears with