Harsh Jaiswal, one of the three researchers behind the Hacktron AI team that chained together two vulnerabilities to compromise OpenAI employees’ ChatGPT and Codex accounts and gain access to OpenAI’s internal code repositories a couple of months ago, has now landed a much bigger payday from Meta for a strikingly similar find.
According to a bounty confirmation email Jaiswal shared, Meta has awarded him a base bounty of $100,000 for reporting a remote code execution vulnerability on Facebook and Instagram that was triggered through HEIC image uploads. The report notes a memory corruption issue in Meta’s image processing pipeline, specifically in how HEIC/HEIF files are handled during server-side image conversion, which could lead to out-of-bounds read/write and potential remote code execution under certain conditions. Meta says it has since fixed the issue, and paid out through its bug bounty program on Bugcrowd. Because Jaiswal sits in the Platinum league of Meta’s Hacker Plus program, he also picked up a 1.15x multiplier, adding a $15,000 bonus on top, for a total payout of $115,000.
The bug class here will look familiar to anyone who followed the OpenAI story from a few weeks ago. In that case, Jaiswal, along with fellow Hacktron AI researchers Mohan Pedhapati and Rahul Maini, found that OpenAI’s community help forum, community.openai.com, could be compromised through a maliciously crafted HEIC image upload. That flaw sat in the ImageMagick library that a decade-old xkcd comic once joked was “holding up civilization”, specifically in libheif, the component ImageMagick calls on to decode HEIC/HEIF files. A heap buffer overflow there let the researchers run code on OpenAI’s forum server, which they then chained with a single sign-on flaw to take over OpenAI employees’ ChatGPT and Codex accounts, eventually opening a proof-of-concept pull request inside OpenAI’s private GitHub monorepo.
The Meta bounty suggests that hunt didn’t stop at OpenAI. Hacktron has described its OpenAI work as part of a broader research effort tracking the same libheif vulnerability class across major tech platforms over a couple of months, with Slack, GitHub Enterprise, and frameworks like Next.js all reportedly probed as part of the same campaign. Meta’s Facebook and Instagram now join that list, and the size of the payout says a lot about how differently companies price out the same underlying bug. OpenAI’s reward for the account-takeover chain that reached its internal repository came to just $6,500, a figure that drew plenty of criticism online given what the researchers were able to access. Meta’s $100,000 base award for what appears to be a comparable memory-corruption bug in image handling is more than fifteen times that, even before the Hacker Plus multiplier is factored in.
Jaiswal, who has been on HackerOne since 2016 under the handle rootxharsh, has more than a decade of bug bounty and vulnerability research experience behind him, including past work with Project Discovery, Zomato, and Cure53, and past finds on platforms like Apple, PayPal, and GitHub. The Meta payout adds one more high-profile name to that list, and reinforces a pattern that’s becoming hard to ignore: a single class of image-parsing bugs, quietly present across some of the biggest platforms on the internet, is turning out to be worth a lot more to some companies than others.