Drew (@bugfireIO) header
Drew (@bugfireIO) profile picture

malware detection and analysis, hunting and gathering, threat research. Views are my own. https://teapaw.org/t.co/efJDIXnaLi

United States
Joined August 2012
Photo by Drew (@bugfireIO)Photo by Drew (@bugfireIO)Photo by Drew (@bugfireIO)Photo by Drew (@bugfireIO)Photo by Drew (@bugfireIO)Photo by Drew (@bugfireIO)Photo by Drew (@bugfireIO)Photo by Drew (@bugfireIO)Photo by Drew (@bugfireIO)
I welcome your typos, your clunky phrasing, your poor grammar, your weird lingo, your odd vocab, arcane vernacular, and inconsistent story telling, I welcome and embrace and appreciate your humanity and the real effort you put into your bad human writing.
6
6
54
2,245
Drew retweeted
New Operation Endgame Timer
7
14
6
148
16,711
Local LLM models are getting good at handling the low hanging fruit of infrastructure pivots. This was one shotted by a ~20GB Qwen3.6 on a macbook, connected up to @urlscanio over an MCP and given some basic analysis skills.
7
3
21
1,443
It’s over. We solemnly swear to never be unaware of Cybersecurity Awareness Month ever again.
12
1
78
9,010
This 👇💯!
Pray that the defenders get access to more powerful models than the attackers - or we’re heading into some very dark times.
27
This is exactly why the frontier model companies need to be directing their energies towards the defensive side first and foremost!
CVE-2026-98432 is a good example of something that really concerns me about the future of cybersecurity. ubuntu.com/security/CVE-2026… The vulnerability was introduced in October 2015, more than 11 years ago, by a performance optimization in the Linux TCP/IPv4 networking stack. One of the most fundamental components of an operating system used by millions of servers. Now Anthropic reports that Claude discovered this flaw and developed a working exploit that allows remote code execution as root, without authentication, against a server with an MPTCP listener. anthropic.com/research/disco… Think about what that means. For decades, we’ve been building critical infrastructure out of enormous amounts of complex, memory-unsafe code. Humans wrote it. Other humans reviewed it, tested it and searched for vulnerabilities. All working with the same fundamental limitations: limited time, limited attention, incomplete visibility and an inability to reason about every possible interaction and race condition. And perhaps that’s what protected us to some extent. Attackers had the same limitations as the people developing and defending these systems. A vulnerability could remain undiscovered for 11 years, even in such a widely used component. Now AI is changing that balance. The ability to inspect code, identify subtle weaknesses and develop working exploits is becoming less dependent on a small number of highly skilled researchers. These activities can increasingly be scaled with compute, automated testing and tokens. And that’s what worries me about the next few years. We’ve accumulated decades of complex, memory-unsafe code in operating systems, network services, applications and libraries. Much of it has never been assessed with anything close to the capabilities that are now becoming available. We may have massively overestimated the security of our software simply because the people looking for vulnerabilities were as limited as the people who introduced them. That balance is breaking down, and I don’t think we’re prepared for what follows.
32
Spent the last few days wishing that DnSpy had an MCP with debugger support, and turns out LLM's can just build that in ~20 minutes if you give it the github repo and a decent spec on what you want. Opus was largely able to one shot the entire mod, and GLM5.3 was able to drive it almost instantly without needing to add any skills. The whole concept of vibe modding an existing tool never really crossed my mind before, but now i'm all here for it.
3
6
1
32
2,402
The latest episode of Behind the Binary is live! We’re diving into everyone's favorite topic—binary similarity—joined by Jonas Wagner and Endre Bangerter from @threatray. 🎧 open.spotify.com/episode/2nm…
2
13
800
In addition to dropping long form deep research on threats or highlighting novel attack chains/campaigns I think its important to remember that threat actors are regularly leveraging known tools. As such its easy to lose sight that an intrusion set doesn't have to be comprised of novel/sophisticated elements to be worth monitoring and proactively protecting your org from. That is part of what inspired the Threat Overview series. Previously I've dropped primers on AsyncRAT, Remcos, and others. Today's drop is about Mythic C2. censys.com/blog/mythic-c2/ Check it out!
1
7
19
1,036
This is important. This isn't just distillation. This is fraud. @Kimi_Moonshot (aka GTG-16002)
holy shit kimi was faking it’s performance by serving you claude you have been using claude all along
1
2
2
8
2,420
Great info here from @craiu !
Wrote a short post on my experience running GLM-5.3-Flash, DeepSeek 4 Flash, and Qwen-3.8-Next locally. Which is the best all-around local model for CTI workflows? medium.com/@costin.raiu/top-…
1
96
Drew retweeted
Wrote a short post on my experience running GLM-5.3-Flash, DeepSeek 4 Flash, and Qwen-3.8-Next locally. Which is the best all-around local model for CTI workflows? medium.com/@costin.raiu/top-…
2
16
2
81
9,130
Drew retweeted
Do you like local AI models? The model that is best for Joe may not be the best for you. My totally unscientific top 10 based on evaluating the quorum on cyber threat intel workflows from last week between various models.
6
8
49
4,085
The recording of "Deobfuscation in the Age of Agentic Reverse Engineering" is now public: youtube.com/watch?v=3-gJ6EUF… We (CC @nicolodev) show how to use agents to break protections found in anti-cheats, DRM systems & commercial protectors. Slides: synthesis.to/presentations/r…
4
104
1
378
28,272
We just put out some new research on #PackClient, a #RAT framework that's being marketed to Chinese-speaking threat actors and is being used by at least one of the Chinese-speaking actors we track. 😎
Proofpoint has observed a Chinese-speaking threat actor (TA4922) using a command and control (C2) framework called PackClient. The framework enables data theft, surveillance, and downloading of additional plugins and payloads. Blog: proofpoint.com/us/blog/threa… #phishing #emailrisk
4
11
892
Drew retweeted
The timer for #flareon13 started on flare-on.com I will likely participate, but not sure I will try to speed-solve it like previous years. Last year was already somewhat pay-to-win. I expect this year to be worse. I don't see any fun in competing against unlimited AI
1
9
61
4,333
If you have a malicious sample in front of you during an active intrusion, spend 20-30 minutes on it. I know the pull to dodge it - the timeline is on fire, EDR is throwing process trees at you, and grabbing the hash and outbound connections feels like enough. But even where it recorded cleanly, process data only shows what ran on that host, at that moment, under those conditions. The binary tells you what to go look for on disk, and half the time it's still sitting there (trust). Thirty minutes of triage and the pieces start fitting together: - Backup C2s that were never contacted because the primary was still up - Drop paths and artifacts the sample writes that never generated an alert, but are still on the host if you know the filename - Config strings and beacon patterns that tie this intrusion to prior activity - Sandbox checks that explain why your detonation came back clean - Conditional persistence
5
33
2
200
10,921
Drew retweeted
One of the coolest Reverse Engineering TTPs. Analyzing APT29's recurring phishing methodologies and provides practical, technical demonstration on how reverse engineers extracts and reverse engineered Commercial C2 to uncover its operational security flaws. Re by @ippsec
1
118
3
641
27,047
Unbelievably embarrassing for Google that Gemini hasn’t committed any cybercrimes yet
256
1,141
149
18,042
833,041
This is an interesting trend worth mentioning. People are processing threat intelligence blogs using AI and republishing summaries and often improved analysis with additional IOCs and YARA rules. These are not always good. 🧵
8
23
2
138
26,846