Pinned Tweet
Announcing my rejected talk for @bsidespyongyang:
Is that "web3 enthusiast" follower with the anime PFP acting suspicious?
Find out if you have confirmed North Korean followers at dprkdetector.app.
Anecdotally I noticed that one of my agents tried to resolve crates through rsproxy.cn which seemed extremely sketchy but is apparently legitimate.
I have .ru and .cn blanket blocked just in case.
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
noid retweeted
Social engineers pretending to be Google Support keep calling me, so I took the opportunity to create this educational video showcasing their tactics.
noid retweeted
🇰🇵
For a few months, I posed as a recruiter and engaged a DPRK persona known as Jason Kwon, who appeared to be operating as hightechrun shortly after news of m00n was shared. He later introduced himself as Roy Hensley. At the present, there is a possibility his current alias is on a shared upwork profile. Without knowing when that profile was put into play, it’s unconfirmed if that alias has entered any projects or been exposed.
Some of this will be familiar to researchers who follow DPRK activity as this wasn’t attribution built from scratch. It started with existing research from @zachxbt being publicly shared and some tracking by @intell_on_chain (@rugpullfinder). During our conversations and conveniently in the Beanstalk discord, he shared the 0xm00n GitHub account and later explained it as an “agency” account.
I used decoy documents, and CanaryTokens to collect what was possible about working environments. The summary covers the identity changes I could follow and connections collected during the small engagement.
I’m publishing this now because I want to get back into documenting, not leaving helpful info sitting in DMs, screenshots, and drafts. Additional context can help someone else connect an account, recognize a recycled persona, or pick up a thread you couldn’t follow further.
Link to medium article ⬇️ and thanks to all involved in this. @0xfigo has recently shared some work done behind the scenes with @north_scan involving tracking DPRK so I’m mentioning that as a resource available for those doing work on our beloved IT Workers.
medium.com/@0xsaiyangod/meet…
Hackers do not care about your timeline or policy
Restrictions only cripple the defenders. It doesn't buy any time. Any hacker who wants to do AI-powered vulnerability research is already doing it.
There is no putting the genie back in this bottle.
I am okay with Kimi 3.1 deliberately leaving out top-end offensive cyber capabilities from their training data.
Discriminating against use cases is MUCH less bad than discriminating against users. And the world does need all the time it can get to prepare, harden, check old code, (properly) formally verify new code, etc before the full weight of very powerful cyber attacks is upon us.
And if this first big wave of AI risk is not handled well, then we may well get a dystopia of highly locked down models, KYC gating on everything, highly intrusive surveillance on everything, a society where you may be a second or third or fourth-class citizen in terms of model access based on your credentials and nationality.
Attackers are starting to abuse XSS / content injection vectors like this - just because something links to the real domain doesn't mean it's legitimate.
Watch out for stuff with big encoded parameters. It's not an exact science but I always remove tracking params out of habit.
The URL: portfolio.metamask.io/?notif……
The parameter is an lz-string compressed JSON object in the Contentful format the Portfolio frontend uses for its own product announcements. The frontend rendered it with the same components as a genuine notification.
Another example.
Heads up - someone is selling alleged @Ledger WalletConnect XSS vulnerability.
Be careful what you click on, an attacker can use a legitimate subdomain to build a credential harvester.
Unverified, but the seller is asking 125k. Consider that when a bug bounty offers you $500.
Watch out for this scam - there are fake crypto job sites. When you get a call back, they try and walk you through sharing your screen, opening terminal, or installing malware.
Hear for yourself what it might sound like!
Crimes on Ethereum Mainnet are also forked to Pulsechain as well, so essentially you get double the charges
noid retweeted
“We used our models to understand the device”
We discovered a sophisticated recording device that was hidden inside one of our offices. We used our models to understand the device and where it had come from, identifying that a specific employee had planted it a few days earlier. We referred the matter to the FBI, and we’re grateful to the agents and prosecutors who moved quickly to make an arrest. The FBI’s investigation is ongoing and we will continue to support their work.
At this time, we have no indications of this individual inappropriately accessing or exfiltrating model weights, source code, or customer data. The employee was terminated.
We take protecting our technology, data, and people seriously, and as a matter of national importance. Our security program is designed to protect against a wide range of threat actors, from insider risks to cybercriminals to nation states, and this case is an example of how AI can help address these sophisticated security threats.
