Researchers Breach OpenAI Internal Repo via HEIF Image Flaw and Overprivileged SSO Token

·8 views

Security researchers disclosed a critical vulnerability chain, assigned CVE-2026-32882, that allowed remote code execution on OpenAI's Discourse-based forum by uploading a crafted HEIF image that exploited a heap overflow in libheif 1.19.7 via ImageMagick. The RCE enabled attackers to harvest an overprivileged OpenAI SSO token from the compromised forum environment, which was then used to access an employee's ChatGPT and Codex account. Through the employee account's connected GitHub integration, researchers demonstrated the impact by having Codex submit a pull request to an internal OpenAI repository. The exploit was developed with assistance from Claude Opus AI models, though the process required expert human guidance and was not fully autonomous. Recommended mitigations include patching libheif, sandboxing image processing, and restricting SSO token scopes to least-privilege principles.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How PostgreSQL's NOTIFY Queue Fills Up and How to Clear It

A developer reproduced the PostgreSQL 'too many notifications in the NOTIFY queue' error using PostgreSQL 17, Docker, and a deliberately limited queue size of 512 KiB set via the max_notify_queue_pages parameter. With a listener session held idle inside an open transaction, 100 large pg_notify() calls were issued from a second connection, filling the queue entirely after 64 successful sends and causing 36 to fail. The key finding is that an open transaction on the listener side blocks notification cleanup, regardless of how large the queue is configured. Simply committing the idle transaction allowed all pending notifications to be delivered and the queue usage returned to zero. This demonstrates that raising max_notify_queue_pages only delays the problem — the real fix is ensuring listener transactions do not remain open unnecessarily.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds Sanity-Powered Repost Marketplace Showcase Without a Full Backend

A developer built Pulse Wall, a content-driven spin-off of Pulse — a social promotion marketplace where businesses run repost campaigns and promoters earn money — using Sanity as the content layer instead of a traditional backend. The project was submitted to the Sanity Challenge under Path Two and reuses an existing Next.js UI with Tailwind and shadcn/ui components, replacing backend services like authentication, payments, and a Go API with Sanity's Content Lake and Studio. Campaign listings and promoter spotlights are managed entirely through Sanity Studio, embedded directly into the app, allowing content updates without redeployment. Key technical hurdles included resolving CORS errors for Studio API calls and fixing a Next.js static prerendering issue where new Sanity content wasn't appearing live until dynamic rendering was enforced in the correct file. The app is deployed on Render with a GitHub Actions workflow pinging the free-tier instance every five minutes to prevent it from spinning down.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

astro-tasks Unifies GitHub, WakaTime, and Git Checks in One CLI Command

A developer has released astro-tasks, an open-source Python tool that consolidates common pre-work checks into a single terminal command. The tool aggregates GitHub notifications, open pull requests, WakaTime coding statistics for the past seven days, and a local repository health scan. It can be installed via pip and offers subcommands to run each check individually, as well as a JSON output flag for piping data into other tools. The utility requires GitHub CLI authentication and a WakaTime configuration file, and supports Python 3.8 and above. Licensed under MIT, the project is hosted on GitHub and the author is actively seeking feedback on its JSON schema and multi-machine WakaTime support.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer builds open log to track AI quota resets and pricing changes across providers

A developer frustrated by undocumented shifts in AI quota reset schedules built Alca, a read-only, no-signup website that tracks model access, pricing, free tiers, and rate limits across AI providers. The site's most-used feature is a dated reset log linking every announced change back to its original source. All content is stored as TypeScript files in a repository, with pull requests serving as the editorial process and CI checks enforcing data integrity — eliminating the need for a database or admin panel. Each data entry carries a verification status indicating whether a source link has been confirmed, making data reliability transparent to users. The site is live at alca-navy.vercel.app and offers a public JSON API under /api/v1 for programmatic access.

0 comments Read more at DEV Community