Intigriti Bug Bytes #240 - September 2026 🚀

Intigriti published Bug Bytes #240 on September 25, 2026, highlighting research on compromising OpenAI, Slack, and Meta through a vulnerable image library and on hacking OpenAI employee accounts in under 72 hours, alongside a $100K Google GFile breach and techniques for attacking AI customer service agents. The newsletter also announced the public beta of CrowdRecon, a crowd-led reconnaissance product, and reported that 79 hackers solved August's Bad Reception CTF with 21 write-ups submitted, while the 0926 Critter Gallery challenge runs until Monday, September 28 for €400 in swag prizes.

Intigriti Bug Bytes 240 - September 2026 🚀 By Ayoub September 25, 2026 Hi hackers, Welcome to the latest edition of Bug Bytes In this month's issue, we'll be featuring: - Compromising OpenAI, Slack, Meta, and more via a vulnerable image library - Hacking OpenAI employee accounts in under 72 hours - Breaking into Google's GFile for $100K - Hacking AI CX agents - Turbo Intruder 2 surpassing 100K requests per second over HTTP/3 And so much more Let's dive in Reconnaissance unleashed: meet CrowdRecon CrowdRecon is officially here https://www.intigriti.com/blog/news/reconnaissance-unleashed-meet-crowdrecon CrowdRecon brings crowd-led context to the moments before and between vulnerability reports. It helps customers quickly identify where attention is needed across exposure, scope, coverage, investigation, and follow-up. For researchers, it introduces new ways to collaborate, contribute, and earn recognition for the valuable recon work that usually disappears. The public beta is open Register today https://www.intigriti.com/researchers/product/crowdrecon and be among the first to try it. Reconnaissance unleashed: meet CrowdRecon How AI changed my day as a QA Engineer In our latest post, Senior QA Engineer Martin Klimovski https://www.linkedin.com/in/martin-klimovski-9975b2b5/ shares how he is leveraging AI for repetitive tasks https://www.intigriti.com/blog/business-insights/how-ai-changed-my-day-as-a-qa-engineer to free up time for deep exploratory work and security testing. The piece covers how he removes friction while keeping judgment, and his views on AI security and safety. From sceptic to supercharged. How AI changed my day as a QA Engineer How AI has changed the way I think, build, and work Senior Software Engineer Koen Van Hauwe https://www.linkedin.com/in/koenvanhauwe/ shares an honest look at a day in his life and his shift from writing code to steering AI https://www.intigriti.com/blog/business-insights/how-ai-has-changed-the-way-i-think-build-and-work . The post covers the Orchestrator shift and the critical role of human oversight, the risks of speed and subtle mistakes that build up over time, and how AI introduces new attack surfaces alongside opportunities for defenders. How AI has changed the way I think, build, and work. A day in the life of an Intigriti Engineer Quick Intigriti 0926 Critter Gallery Challenge is still live Intigriti's 0926 Challenge, Critter Gallery by @khanhdlq https://x.com/khanhdlq , is still ongoing. Capture the flag before Monday the 28th of September for a chance to win €400 in swag prizes. Intigriti Challenge 0926 Intigriti 0826 Bad Reception CTF results are in August's CTF challenge, Bad Reception, featured a broken TV that revealed the flag once fixed. This challenge definitely sent lots of participants down rabbit holes and dead ends, which made it more engaging than ever. Quick recap: - 79 hackers found the correct solution - 21 hackers wrote a cool write-up If you want to put your hacking skills to the test, be sure to give Bad Reception 0826 https://challenge-0826.challenges.intigriti.io/ a go before heading over to Bugology https://bugology.intigriti.io/intigriti-monthly-challenges/0826 , where you can find all the researchers' submitted solutions. Intigriti Challenge 0826 Blogs & videos Hacking AI customer service agents https://www.intigriti.com/researchers/blog/hacking-tools/hacking-ai-customer-service-agents Hacking AI customer service agents Cover Image Most support chatbots have evolved into fully autonomous agents that can help with almost anything, and that comes with additional risk. Our latest article, based on @intidc https://x.com/intidc 's talk at Bug Bounty Village during DEF CON 34, breaks down the full attack surface in AI customer service agents https://www.intigriti.com/researchers/blog/hacking-tools/hacking-ai-customer-service-agents : from tricking them into leaking sensitive data via email spoofing to invoking unauthorized tool calls on behalf of the victim. All without ever touching an automated scanner or proxy interceptor. - Looking for a complete guide on file upload vulnerabilities? Our full guide on exploiting insecure file uploads https://www.intigriti.com/researchers/blog/hacking-tools/insecure-file-uploads-a-complete-guide-to-finding-advanced-file-upload-vulnerabilities covers everything from basic bypass techniques to advanced exploitation scenarios. - Reports sometimes get closed incorrectly, downgraded in severity, or left pending for longer than expected. When that happens, knowing how to respond through the right channels matters. Our comprehensive guide walks you through Intigriti's mediation process https://www.intigriti.com/researchers/blog/hacking-tools/how-to-appeal-a-bug-bounty-submission , the most common scenarios researchers face, and the mistakes to avoid during the process, such as unauthorized public disclosure, which can work against you rather than help your case. - Going from zero to your first valid bug report? In case you missed it, we recently launched the Bug Bounty Starter Kit https://www.intigriti.com/bug-bounty-starter-kit , a free guide covering everything from recon and tooling to the exploitation of SQLi, XSS, and BAC vulnerabilities, to finally how to learn to write a compelling vulnerability report that gets triaged faster. Get your copy now. https://www.intigriti.com/bug-bounty-starter-kit Tools & resources Tools GeminiHunter https://github.com/devploit/geminiHunter GeminiHunter Did you know that some exposed Google Maps API keys also have access to Gemini models? GeminiHunter https://github.com/devploit/geminiHunter by @devploit https://x.com/devploit finds and validates exposed Google Gemini API keys across web assets, source maps, Wayback snapshots, and Android APKs. It deduplicates findings, checks if keys are actually valid, and attempts bypass variations on restricted ones. If you are hunting Google API key exposures, this adds a new dimension to check. - Prompt injection is the top risk in the OWASP LLM Top 10. Awesome Prompt Injection https://github.com/Joe-B-Security/awesome-prompt-injection is a curated collection of resources covering everything from research papers and real-world attack chains to hands-on CTFs and detection tools for testing LLM and agentic applications. - Want your AI agent to interact directly with Chrome? Chrome DevTools MCP https://github.com/ChromeDevTools/chrome-devtools-mcp lets AI agents use Chrome's debugging protocol to inspect network requests, DOM elements, execute JavaScript in the console, and capture screenshots, all programmatically. Resources HEIF Heist https://x.com/rootxharsh/status/2100801820960620574 HEIF Heist research @rootxharsh https://x.com/rootxharsh and team published HEIF Heist https://x.com/rootxharsh/status/2100801820960620574 , a months-long investigation into libheif that compromised OpenAI, Slack, Meta, GitHub Enterprise, Rails, Next.js, ImageMagick, and more. The attack surface was a single obscure image parsing library sitting beneath a huge number of applications. Some of the RCE and information leak attempts only landed after thousands of image uploads, and only one organization caught the exploitation in progress. - Two bugs, 72 hours, and a PR in OpenAI's internal codebase. @S1r1u5 https://x.com/S1r1u5 and team published the details on how they took over ChatGPT and Codex accounts of OpenAI employees https://x.com/S1r1u5 /status/2100777801335095383 , reaching connected services including Outlook, Slack, and GitHub. - Breaking into Google's GFile for $100K. Brutecat documents how he leveraged AI to scan discovery documents at scale across Google's infrastructure https://bughunters.google.com/blog/breaking-into-googles-gfile-for-100k , then found a new approach to exploit the GFile library for access to internal filesystems and storage. - A valid URL with a Command in its Shadow. @saur1n https://x.com/saur1n found that a connectivity check was passing user input to cURL inside a shell, and so was a second feature which ultimately led to a command injection vulnerability https://saurinn.github.io/blog/from-input-to-shell/ . - Turbo Intruder 2 has landed. @t0xodile https://x.com/t0xodile from PortSwigger released a new version of Turbo Intruder https://portswigger.net/research/http3-in-burp-suite with a native HTTP/3 engine, surpassing 100,000 requests per second over WiFi. It also includes a Burp adapter for HTTP/3 exclusive targets and new research-grade race condition techniques. - Full control over 3,000 companies and all their registered branches. @stuipds https://x.com/stuipds published a detailed writeup covering three weeks of studying a target to find a chain of vulnerabilities that led to complete company takeover via secondary context. https://medium.com/@Berserker1337/secondary-context-leads-to-company-takeover-0dc88ca751ea - A remote kernel vulnerability in macOS SMBFS, $20,000 bounty from Apple. @slinafirinne https://x.com/slinafirinne published the writeup for CVE-2026-84543 https://github.com/petermalone/CVE-2026-84543 , his second $20K Apple bounty. - Executing an ELF without touching the filesystem. This research article demonstrates using the Linux kernel keyring to stage an ELF in slab memory https://matheuzsecurity.github.io/hacking/linux-kernel-keyring-fileless-exec/ and execute it via userland exec, skipping execve and the filesystem entirely. - Exploiting tool and function calling in LLM agents. Sentry Security published a post covering the attack surface in LLM agents https://blog.sentry.security/exploiting-tool-and-function-calling-in-llm-agents/ when they are given access to external tools. - Path traversal plus arbitrary file write turned into RCE. @MrTuxRacer https://x.com/MrTuxRacer published the details on CVE-2026-28373 in the Stackfield desktop app https://www.rcesecurity.com/2026/03/stackfield-desktop-app-rce-via-path-traversal-and-arbitrary-file-write-cve-2026-28373/ , where an encrypted backup was enough to achieve remote code execution. - LFI filter bypass via emojis. @0xConda https://x.com/0xConda shared a practical path traversal bug bounty tip https://x.com/0xConda/status/2101647631327940732 with emojis to bypass strict filters. - @0xacb https://x.com/0xacb shares a practical bug bounty tip on fundamental concept of parser differentials https://x.com/0xacb/status/2067535207935131916 . Understanding the concept in depth can help you land more vulnerabilities, consistently. - To read private chat rooms in Better Messages, just tell it you are the AI bot. @ CryptoCat https://x.com/ CryptoCat published the analysis of CVE-2026-89093 https://cryptocat.me/blog/research/analysis/cve 2026 89093/?420 : identity was an IP-prefix check, and the IP came from a request header. - Guest checkout coordinates flowing straight into a SQL query. @ CryptoCat https://x.com/ CryptoCat also published another writeup on CVE-2026-18442 in WCFM Marketplace https://cryptocat.me/blog/research/analysis/cve 2026 18442/ , where a store-distance shipping calculation was vulnerable to SQL injection. - How close is "one click away from shutting down every restaurant in Europe"? This writeup covers a high-severity issue in a restaurant platform that allowed malicious users to disrupt the delivery service. https://medium.com/@TionoX/how-i-could-have-shut-down-every-restaurant-in-europe-with-one-click-0125387c1628 - Two researchers bought cheap domains including noreply.net and deleteduser.com, then set up email listeners. Hundreds of companies have been caught sending corporate secrets to those emails. Read the full story covered by Wired. https://www.wired.com/story/sensitive-info-goes-into-no-reply-emails-constantly-this-guy-sees-it-all/ - $76,000 from a single program on Bugcrowd. @4non Hunter https://x.com/4non Hunter published a breakdown of how sustained focus https://anonhunter.medium.com/how-i-earned-76-000-from-a-single-program-on-bugcrowd-adf2a0eeece0 on one target led to a significant payout. - A year of hacking with LLMs. @edwardzpeng https://x.com/edwardzpeng shared his keynote from the Offbyone security conference, covering a full year of integrating LLMs into vulnerability research https://sites.google.com/site/zhiniangpeng/blogs/Hacking-with-LLMs-Eng . - Rickrolling the entire FIFA World Cup. This writeup details how a single ID was enough to gain unauthorized control over an administrative panel https://bobdahacker.com/blog/fifa-hack that could in theory have let him rickroll the entire FIFA World Cup. - Escalating a blind SSRF to internal cloud metadata, $4,200 bounty. This researcher documented how he was able to escalate a blind SSRF to read cloud metadata https://medium.com/@t4nv1/4-200-bug-bounty-escalating-blind-ssrf-to-internal-cloud-metadata-credential-exfiltration-987f3c0ab032 . - An IDOR in an invite cancellation flow exposed personal data. The writeup covers how scheduled, recurring actions led to unintended PII exposure https://medium.com/@ma6632174/i-wasnt-looking-for-pii-i-just-cancelled-an-invite-idor-pii-41d1df546edf . - A default password let one researcher log into almost anyone's account. The writeup covers a default-credential issue https://medium.com/p/75d3a60dbc5b leading to mass account access. - 11,000+ bug bounty reports in one searchable library. @ Shark byte https://x.com/ Shark byte compiled an open library of disclosed reports across platforms. https://bug-bounty-disclosures.vercel.app/ Whether you're new to web hacking or a well-seasoned researcher, you can always learn from other researchers' findings and experiences. - Mutating Safari. @garethheyes https://x.com/garethheyes published a brief post on a Safari-specific behavior https://thespanner.co.uk/mutating-safari found by Shazzer. - Pwnfox, but for Brave. Now that Brave supports containers, @leo rac https://x.com/leo rac built pwnbrave https://github.com/leorac/pwnbrave to replicate the Pwnfox workflow without being tied to Firefox. - Built to pass OSCP, now used by the community. @DotNetRussell https://x.com/DotNetRussell built BloodBash https://github.com/SquidSec/BloodBash for Active Directory practice and shared it at DEF CON 34, where others told him they used it to pass their OSCP as well. - James Kettle's @albinowax https://x.com/albinowax ' Can AI do novel security research? https://www.youtube.com/watch?v=jCFZFDHnZrQ ' talk is now on YouTube - Did OpenClaw really hack a gym? @PinkDraconian https://x.com/PinkDraconian published a video investigating whether OpenClaw really hacked a gym https://www.youtube.com/watch?v=3rbZO2WTNEI to claim an already claimed spot. Company news Wild West Hackin' Fest Intigriti is a Gold Sponsor of Wild West Hackin Fest in Deadwood, South Dakota. If you are attending, be sure to come say hi Wild West Hackin' Fest Feedback & suggestions Before you click away: Do you have feedback, or would you like your technical content to get featured in the next Bug Bytes issue? We want to hear from you. Feel free to send us an email at community@intigriti.com mailto:community@intigriti.com or DM https://twitter.com/messages/compose?recipient id=736615159653924864 us on X/Twitter, and we'll take it from there. Did you like this Bug Bytes issue? Consider sharing it with your friends and tagging us along on X/Twitter, Instagram, or LinkedIn. Wishing you a bountiful month ahead, Keep on rocking Author Ayoub Senior security content developer You may also like August 28, 2026 Intigriti Bug Bytes 239 - August 2026 🚀 Welcome to the latest edition of Bug Bytes In this month's issue, we are featuring: Intigriti as the new provider for Adobe's Bug Bounty Program CSS injection as an attack vector inside your email inbox AI doing novel security research: the HTTP Terminator 169 offensive recon skills in one AI-ready Read more: Intigriti Bug Bytes 239 - August 2026 🚀 https://www.intigriti.com/researchers/blog/bug-bytes/intigriti-bug-bytes-239-august-2026 July 31, 2026 Intigriti Bug Bytes 238 - July 2026 🚀 Welcome to the latest edition of Bug Bytes In this month's issue, we'll be featuring: Intigriti turns 10 RCE in GitHub.com and GitHub Enterprise Server Burp Suite going agentic with Burp AT Hacking Gemini Enterprise for $15,000 3,708 live credentials found by scanning GitHub Archive And so much mo Read more: Intigriti Bug Bytes 238 - July 2026 🚀 https://www.intigriti.com/researchers/blog/bug-bytes/intigriti-bug-bytes-238-july-2026 June 26, 2026 Intigriti Bug Bytes 237 - June 2026 🚀 Welcome to the latest edition of Bug Bytes In this month's issue, we are featuring: A 10-year-old pre-auth RCE in phpBB Earning $500K hacking Google with AI Reading any Salesforce Marketing Cloud account's emails New DOMPurify sanitizer bypass Mapping abandoned S3 buckets to redo SolarWinds at scal Read more: Intigriti Bug Bytes 237 - June 2026 🚀 https://www.intigriti.com/researchers/blog/bug-bytes/intigriti-bug-bytes-237-june-2026