Andrius Useckas

Breaking AI agents before the bad guys do. CTO @ ZioSec. Founded ThreatX (acq. A10). 25+ yrs offensive security. Dad, snowboarder, biker. Opinions happen.

[Andrius Useckas @daronin]

@JakeKing Some are four security researchers in hoodies. Know the difference. :)

[Andrius Useckas @daronin]

@AndrewYang It’s not just cost. Availability matters too. I can tolerate downtime when using a model for coding. I’m much less willing to build an enterprise product around a service I can’t count on. Frontier capabilities don’t automatically mean enterprise reliability.

[Andrius Useckas @daronin]

@hetmehtaa Yea, and even if you enroll Yubikeys, the process gets stuck on verification.

[Andrius Useckas @daronin]

@CristiVlad25 @OpenAI @thsottiaux @romainhuet Yeah, the process is broken. I’ve been through three different flows and submitted my docs 5+ times. Still no Daybreak access. “We want to work with outside experts,” but won’t let them in. Feels like gaslighting at this point.

[Andrius Useckas @daronin]

@UK_Daniel_Card Knowing how much they invest in cybersecurity down under, I can just picture someone spotting OpenAI crawler in the logs and going: "OMG, we’ve been hacked by AI."

[Andrius Useckas @daronin]

@AndrewCurran_ I don’t know, maybe invest in cybersecurity? Most hackers probably weren’t that interested in Medicare statistics. Apparently the agent was.

[Andrius Useckas @daronin]

@ZackKorman I doubt it had to try that hard, knowing the state of cybersecurity down under.

[Andrius Useckas @daronin]

@_jensec @OpenAI Yup, verification appears to be broken. Good thing I had existing yubikeys.

[Andrius Useckas @daronin]

Forget AI jumping air gaps via CPU heat. The real threat is old-school attacks at a new scale, powered by AI orchestration. Sorry, SQLi protection still matters.
https://t.co/LFKWdXFI5h

[Andrius Useckas @daronin]

Busy day for frontier model releases. Guess we’re not "pacing the frontier" after all.

[OpenAI @OpenAI]

Please welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.

GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.

We’ve also made caching and inference more efficient, and we’re passing the savings directly to you: 50% lower API prices for Sol and Luna compared with GPT‑5.6 promotional pricing.

[Andrius Useckas @daronin]

@HackingDave Agreed. It no longer sounds like a drunken PhD, and it doesn’t burn through your entire session in half an hour like Fable 5.1.

[Andrius Useckas @daronin]

@AnthropicAI Per Anthropic’s own announcement, they’re funding their evaluator. Can Accenture publish findings Anthropic doesn’t like without needing their approval or risking that funding? That’s what I’d want to know before calling this independent.

[Andrius Useckas @daronin]

A take on the "AI apocalypse" from cybersecurity professionals' perspective: https://t.co/1G3vvQtGgB

[Andrius Useckas @daronin]

@AlpinDale Well, it’s a long game, obviously. That’s only about half a lap around the galaxy.

[Andrius Useckas @daronin]

If we ever create sentient AI (a big if), what philosophy should we pass on to it? Intelligence alone doesn’t make us good or bad. Some AI labs already employ philosophers. I’m more interested in whose values they’re teaching than another prediction of doom or utopia.

[Andrius Useckas @daronin]

@ZackKorman The surprising part is seeing otherwise serious technical people push these edge cases as if cybersecurity has never considered them. At some point, you start questioning the motivation.

[Andrius Useckas @daronin]

@andreamichi Finally someone speaks reason.

[Andrius Useckas @daronin]

Temperature sensor access can be restricted. The BitWhisper study managed 1–8 bits per hour within 40 cm, with both machines already compromised. The receiver also has to distinguish deliberate signaling from ordinary workload heat. Quite a few prerequisites missing from this story.

[Andrius Useckas @daronin]

@NaveenGRao I don’t think people care about understanding anymore. Just picking a camp and repeating the slogans.

[Andrius Useckas @daronin]

@UK_Daniel_Card Hearing politicians talk about something you actually understand really makes you wonder how much of everything else they say is complete BS.