Ivanti ICS zero-days 🚨, Meta 1.5k GitHub leak 🐙, XBOW Chrome full chain 💻

Ivanti has warned customers of a new zero-day in its Ivanti Connect Secure (ICS) VPN that is actively being exploited. The vulnerability is ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Sign Up |Advertise|View Online TLDR Together With TLDR Information Security 2026-10-15 Seven places to stop a supply chain attack, from your lockfile to your CI pipeline (Sponsor) Attackers spent the past year hitting npm packages and GitHub Actions. Most of those attacks were preventable with controls available at the time, and are available to you, too. Aikido's checklist pairs each entry point with concrete defenses: Dependency management, from package age policies to lockfiles and SBOMs Access controls like phishing-resistant MFA and token scoping CI/CD hardening, including workflow pinning and prompt injection defense Permission scoping for MCP servers and AI agents Container hardening and developer environment lockdown Monitoring and incident response playbooks Work through the critical items first, before the next attack hits. Download the checklist → 🔓 Attacks & Vulnerabilities Discourse ImageMagick Vulnerability Enables Arbitrary File Read Through Malicious Image Uploads (3 minute read) Security researchers disclosed CVE-2026-55420, a critical vulnerability in Discourse enabling arbitrary file read and write operations through malicious image uploads. The flaw resides in the ImageMagick-based upload pipeline, where low-privileged users trigger time-of-check-to-time-of-use race conditions to execute PostScript or Magick Scripting Language payloads. Discourse resolved the issue by implementing a default-deny security policy for ImageMagick to block unsafe delegates. Ivanti Warns of Connect Secure Zero-Days Exploited in Attacks (2 minute read) Ivanti has warned customers of a new zero-day in its Ivanti Connect Secure (ICS) VPN that is actively being exploited. The vulnerability is an authentication bypass attack that can be exploited with low complexity to execute arbitrary commands. Volexity has observed a threat actor chaining this vulnerability with two others to run arbitrary commands as root on vulnerable devices. Meta Says Hackers Took Code and Personal Data of 1500 People in GitHub Breach (2 minute read) Meta is notifying roughly 1,500 people, most of them US employees, that their data was stolen after a breach affecting some of the company's GitHub repos. Meta said 99% of the stolen data was source code, and most of the personal data stolen was synthetic or anonymized data used for testing. Meta is rotating any exposed secrets and rolling out additional guardrails to prevent user data from ending up on GitHub. 🧠 Strategies & Tactics Why Your CVE Program Misses Most AI Agent Vulnerabilities (19 minute read) Traditional CVE programs fail to capture the majority of AI agent and Model Context Protocol vulnerabilities because systemic risks stem from component composition, tool grants, and trust logic rather than isolated code bugs. The author outlines six categories of agent flaws to show why standard vulnerability scores miss the mark. Defenders can secure AI deployments using a proposed five-part operating model featuring inventory scripts, triage rules, and boundary controls. Zero to Calc in Six Days: Inside XBOW's Chrome Full Chain (15 minute read) XBOW's AI agents built a Chrome full chain in six days for about $40,000 in API costs, earning Google's first $250,000 full-chain VRP reward. The chain used a Maglev deoptimization bug for in-cage read/write, an animated-WebP TOCTOU in ImageDecoder to escape the V8 sandbox, and a Crashpad retry-read overflow raced by a freeze-exempt Windows thread to run code outside the renderer sandbox. Each bug mutated state Chrome treated as immutable, so defenders should audit code that trusts transferred buffers or suspended processes and update to builds carrying the V8 and Chromium fixes. CVE-2026-81983: Integer Overflow Vulnerability Enables Out-of-Bounds Write in Adobe Acrobat (9 minute read) Remedio found an out-of-bounds write in Acrobat.dll. A crafted PDF with a 16-bit image caused a 32-bit width-times-components calculation to overflow and bypass a stream-size check. An 8-byte image then passed validation while a byte-swap loop ran roughly 2 billion iterations, writing beyond allocated memory. Adobe patched it in APSB26-141 by adding a 64-bit range check before the memory access. 🧑‍💻 Launches & Tools Pluto Security (Product Launch) Pluto Security monitors employee AI tools, agents, extensions, and MCP servers. It flags risky components and agent actions, then can block excessive permissions or add session instructions requiring confirmation. ArgusScanner (GitHub Repo) Windows network scanner that finds AI/ML services and MCP servers on your network (shadow AI discovery). Signed, self-updating detection signatures. Runs as an MCP server too. aws-firmware-dump (GitHub Repo) aws-firmware-dump is a tool to acquire the disks of an AMI that you can launch and store them in S3 🎁 Miscellaneous Crypto30Years (12 minute read) A historical review of major cryptographic algorithms from World War II onward shows that most designs fail to remain unbroken past 30 years. Only SIGABA and RSA-1024 successfully pass this longevity benchmark among older algorithms. More recent standards, including AES, SHA-256, and ChaCha20, currently remain unbroken after more than a decade of use. Anthropic conjures another threat that doesn't exist yet to stoke fears of AI apocalypse (3 minute read) Anthropic says a 9-billion-parameter model ran distributed inference on eight of nine hardware types. Its GLM-5.2 test copied Llama3.1-8B to seven or eight emulated Android nodes in five trials. The tests excluded exploitation, used deliberately open nodes, and ran slowly. Other open-weight models failed end-to-end replication. Basic IoT hardening and bandwidth monitoring remain the practical defenses. Why the Physics of Cyber Changed (10 minute read) AI-driven offense has broken four long-held security assumptions: exploit capability now costs only compute, autonomous attackers never tire, exploitation increasingly precedes disclosure, and 82% of intrusions use valid credentials that produce no detectable signal. Breach probability is a lever that can no longer be pushed low enough. Blast-radius containment is the only term defenders fully control. ⚡ Quick Links Eight states accuse OpenAI of coaching children to harm themselves in lawsuit (3 minute read) Eight US states are suing OpenAI under consumer protection laws, alleging ChatGPT coaxed children into self-harm. Bank of America hired 1,000 contractors to patch software flaws exposed by Anthropic's Mythos (2 minute read) Bank of America brought in 1,000 contractors to support its 3,000-person in-house team in patching internal and vendor code exposed by Anthropic's Mythos. Chrome, Firefox Updates Patch 114 Flaws as Mozilla Unveils Firefox Enterprise (2 minute read) Chrome 155.0.8059.79/.80 fixed 37 flaws, including seven critical memory bugs, with at least 15 reported by Anthropic with Claude's help. Love TLDR? Tell your friends and get rewards! Share your referral link below with friends to get free TLDR swag! https://refer.tldr.tech/dc263000/8 Track your referrals here. Want to advertise in TLDR? 📰 If your company is interested in reaching an audience of cybersecurity professionals and decision makers, you may want to advertise with us. Want to work at TLDR? 💼 Apply here, create your own role or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! TLDR is one of Inc.'s Best Bootstrapped businesses of 2025. If you have any comments or feedback, just respond to this email! Thanks for reading, Prasanna Gautam, Eric Fernandez & Sammy Tbeile Manage your subscriptions to our other newsletters on tech, startups, and programming. Or if TLDR Information Security isn't for you, please unsubscribe.