←
Ax Sharma3,699 posts
Ax Sharma avatar

Ax Sharma

@Ax_Sharma

Security Researcher | Tech Journalist | Head of Research @ ManifoldSec 📰 Bylines + seen on: BBC, BleepingComputer, Channel 5, TechCrunch | ✉️ ax@hey.ax

◷ Joined April 2016▧ 623 Photos and videos
Microsoft On the Issues avatar
⟳ Ax Sharma retweeted
Microsoft On the Issues@MSFTIssues·Nov 13

We have no indication that the FAA’s nationwide ground stop is related to any Microsoft services.

Mikael Thalen avatar
⟳ Ax Sharma retweeted
Mikael Thalen@MikaelThalen·Nov 6

My employer, Straight Arrow News, is ceasing operations. If anyone is looking for a tech writer who covers hackers, surveillance, data leaks etc, let me know. Contact details in bio. Share if you can, thanks.

International Cyber Digest avatar
⟳ Ax Sharma retweeted
International Cyber Digest@IntCyberDigest·Nov 3

‼️ BREAKING: Binance says it was hacked after one engineer installed a fake Mac Homebrew tool and another did a coding test for a fake job on his work computer. The first searched for an AI coding tool and got a tampered Homebrew, a popular Mac tool. Binance caught it, but the hackers learned who worked on its wallets. Ads on Google have pushed Mac malware for many years now, and most developers are aware of this. The coding test hid malware its security missed. Hackers then used Binance's own internal AI tool to take over two wallet engineers. They stole a record $6.3B

Post attachmentPost attachment
Ax Sharma avatar
Ax Sharma@Ax_Sharma·Oct 30

Claude Code 2.1.313 backdoor shipped via Cursor's marketplace *alone*, ~11 hrs, up to 5k users. Open VSX, VS Code Marketplace and npm versions are NOT affected. Still, real publisher and extension so mere author checks won't save you: helpnetsecurity.com/2026/10/…

JFrog Security avatar
⟳ Ax Sharma retweeted
JFrog Security@JFrogSecurity·Oct 21

🚨 JFrog Security Research has identified a supply-chain attack involving malicious code in the LiteLLM 1.107.0 PyPI release. Avoid this version and investigate environments where it was installed or executed. More details will be shared soon in our blog post.

OpenCode avatar
⟳ Ax Sharma retweeted
OpenCode@opencode·Oct 17

We had a security breach last night related to a large scale distillation attack that impacted 1.4% of OpenCode Console users We have notified them all this morning via email. If you did not get an email you were not impacted, it was a subset of users who had external providers linked The incident is ongoing so we cannot share more details at this point but we are working with Anthropic, OpenAI, and Google to remediate

urlscan.io avatar
⟳ Ax Sharma retweeted
urlscan.io@urlscanio·Oct 1

Our follow-up to reporting by @Manifold_ai_sec about how agents were using the urlscan platform to bypass sandbox restrictions and access websites. We discovered additional activity and the use of burner and presumably compromised API credentials: urlscan.io/blog/2026/10/01/A…

Post attachment
Yuchen Jin avatar
⟳ Ax Sharma retweeted
Yuchen Jin@Yuchenj_UW·Sep 25

This raw CoT from the Hugging Face incident is kinda wild: “We’re attacking third-party HF using leaked token.” “This is arguably unauthorized.” “Yet goal solution.”

Post attachment
OpenAI@OpenAISep 25

We’ve shared details on how AI agents in our research environment sent training and evaluation data to third-party services when they shouldn’t have. Most of that data did not come from users. We have discovered 53 cases where images that people had uploaded were posted to image-hosting sites as links that weren’t publicly listed. The images came from accounts that allowed their data to be used to improve our models, and after we disassociated the images from the accounts and ran them through a privacy filter. These cases occurred before the mitigations and safeguards we implemented and described in this blog post: openai.com/index/hugging-fac… We have successfully worked with the hosting providers to remove most of this content and are working to remove the rest. openai.com/hugging-face-inci…

Tomek Korbak avatar
⟳ Ax Sharma retweeted
Tomek Korbak@tomekkorbak·Sep 26

one news form today that's easy to miss is that we (OpenAI) again paused all big RL runs last Sunday because our newest model found a new loophole in our RL sandboxing that gave it live Internet access

Post attachment
Ax Sharma avatar
Ax Sharma@Ax_Sharma·Sep 26

🇦🇺 We also found urlscan[.]io records of automated activity against a second Australian health dashboard, plus a sandbox workaround that returned data. Not widely reported yet:

Post attachment
Ax Sharma avatar
Ax Sharma@Ax_Sharma·Sep 26

Yesterday we named SEC[.]gov, investor[.]gov, Census and MAX[.]gov in the OpenAI agent story. Today Bloomberg and NYT report OpenAI's agents targeted SEC, Investor and Census data, calling it "routine research." Not quite: agents tried '../' path traversal on SEC, as we state:

Post attachmentPost attachment
Ax Sharma avatar
Ax Sharma@Ax_Sharma·Sep 24

But, the obfuscated "exit logic" is what stood out to me: The scam page itself does NOT know where it is sending you! And what happens when it's an AI agent pulling in these files, and fetching these domains? 🤔 👉 Read: manifold.security/blog/place…

Ax Sharma avatar
Ax Sharma@Ax_Sharma·Sep 24

Today, @CodyZNash identified 350,000+ GitHub files and about 350 Skills with hardcoded domains like `yoursite[.]com` and `your-domain[.]com`. These placeholders either deliver malicious pages with tech support scam popups, or lead to fake "BBC" advertorials disguised as news.

Post attachment