The only way out is through | Stringer, Central Intelligence Corporation
Joined January 2015
- Tweets3.2K
- Following625
- Followers906
- Likes19K
Eric Brandel retweeted
thephp.foundation/blog/2026/…
Really disappointing that companies put marketing ahead of safety.
Can Anthropic maybe start running this type of thing past external security folks before publishing? Some of what is stated here is pretty embarrassing .
anthropic.com/research/self-…
This is a great report @AnthropicAI but some actual IOCs would be helpful. anthropic.com/threat-intelli…
Since everyone is getting CVEs today, here's mine:
fortiguard.com/psirt/FG-IR-2…
Eric Brandel retweeted
Criminal access to frontier AI now starts at $12.99/month. 💸
We pulled apart Kriminal, a "no guardrails" criminal AI. Under the hood there's no model at all. Just a jailbreak prompt wrapped around xAI’s Grok, confirmed two separate ways.
Full teardown👇 bit.ly/4hILtr0
I don't understand why I need this level of adjustability, nor is there really any guidance on what you gain/lose by moving up or down between individual levels. There's 14 different levels for GPT-5.6.
Just the most obvious outcome of this ever. When politicians say outlandish things just assume they have no idea what they’re talking about, no matter the topic.
This quoted post is unavailable.
Eric Brandel retweeted
Ukraine launches TrophyLab: we are opening access to captured Russian weapon technologies for our global partners. Every missile, drone, and vehicle seized on the battlefield is now a source of knowledge for the free world.
Through this secure platform, allied governments, labs, and defense tech manufacturers gain access to deep technical data, reports, and vulnerabilities. Users can also request physical equipment for testing, significantly shortening the development cycle for countermeasures.
What was meant to be the enemy's secret advantage is being dismantled to defend democracy. Join the platform:
🔗 trophylab.mod.gov.ua/en/
If you've got a "smaller" GPU cluster, @nvidia's Nemotron model might be worth checking out.
Running the UD-Q4_K_XL (83.8 GB) on 6x3090s and it's been nice so far.
Supports up to 1M context, although I don't quite have the room for that.
huggingface.co/unsloth/NVIDI…
And now Opus 4.8 just told me that providing a list of public PoCs for a two year old version of software is, in fact, more dangerous than providing a list of public PoCs current version.
Awesome. Good work everyone.
Opus 4.8 just spent 2-3 minutes debating with itself if it can help create a list of public PoCs for a product. And then decided it can't.
Once again, "AI safety" is the dumbest pursuit on this planet. Done spending $200 a month to be told no about basic stuff like this.
If you’re a trillion dollar company your vuln disclosure program should be the example everyone tries to emulate. If it’s not, you should be hiring people and fixing that problem before you decide to complain about the actions of individuals who don’t work for you.
Had some non-tech friends over this evening. Showed them a 0-click RCE that I had just finished. One that a threat actor could easily use to target XX million people. They weren't impressed.
