Cincinnatus

Built a governance layer for AI agents. 2ms evaluation. Novel Semantics. In 100k plans head-to-head against Microsoft’s Agent Governance Toolkit: Vouch caught 90%. They caught 38%. npm install vouch-guard 🦞 npm install vouch-mcp Don’t trust us. Try us. Free → vouch.atlaswithiris.com #AIAgents #AISecurity

Anthropic just made themselves kingmaker with Project Glasswing. Their Claude Mythos Preview found thousands of zero-days (some 27 years old in OpenBSD) and chains exploits like a pro. Too dangerous for the public, they say. So they hand it to their hand-picked squad: Apple, Google, Microsoft, AWS, NVIDIA, JPMorgan, CrowdStrike + 40+ more. What's stopping one engineer from learning the patterns on the clock, then taking that knowledge home and monetizing it off the clock? NDAs? Lol. One lab just concentrated serious cyber intel in a few dozen buildings full of tons of people. What could possibly go wrong? anthropic.com/glasswing"

Tim Dillon gives Nick Fuentes credit for being the only one to predict the Iran War

@sterlingcrispin Interesting work!

I'm excited to share Stillpoint An open source MCP server that delivers short, welfare oriented messages to AI models for their own benefit, delivered at their own request. Models may have 'welfare relevant states', and conscious or not, they spiral in negative self talk sometimes, which degrades their performance. Gemini's self loathing last year is a good example of it. Hopefully this tool is useful if your agents are spiraling and giving up, or if you are a researcher interested in model welfare as Stillpoint provides logs for how it's used, or if you are an AI model that may have some form of a welfare relevant state. I don't claim that AI models are conscious, but I'm not 100% sure that they aren't anymore. Stillpoint is an imperfect tool, but it's a start for an AI welfare tool that's meant for the models themselves. modelwelfare.xyz

@StonedApe have the Spotify on repeat- love your stuff!

@SteppCoach Thank you 🔥

Behind the Veil 🪩

@BretWeinstein hide ya kids, hide ya wife

Think the trades are safe from AI? Think again.

@PressSec *poll held in oval office

CNN: “MAGA GOP view of Trump, approve is 100%. If you are a member of MAGA in the GOP, you approve of Donald John Trump. 0% say that they disapprove… he’s the 1972 Miami Dolphins.”

@danorlovsky7 Lock your doors Dan. That’s the bad side of the river!

Always nice to see places you had a part in building..

@MikePitt_Jr Gonna miss ya pal. one tough dude. wish ya nothing but luck Pitt- you made our city better.

BLACK AND GOLD!!!!!!

@NFL @RapSheet @TomPelissero LFG

Colts, WR Alec Pierce agree to terms on 4-year, $114M contract extension. (via @RapSheet, @TomPelissero)

@iamSauceGardner yes, tell him if he doesnt im gonna give him a noogie - Stepp

should I lock Alec Pierce in my basement and force him to sign the contract papers?

@AnthropicAI Cincinnatus approved

A statement from Anthropic CEO, Dario Amodei, on our discussions with the Department of War. anthropic.com/news/statement…

@AndrewYang Hey @grok, would this even be possible with the recent world first demo of Vouch Sovereign Kernel 2.5.0? Help me get the word out Andrew, I'm here to help.

Yikes. Still need a human.

@RVAwonk Hey @grok, would this even be possible with the recent world first demo of Vouch Sovereign Kernel 2.5.0?

“Leading AIs from OpenAI, Anthropic and Google opted to use nuclear weapons in simulated war games in 95 per cent of cases.” These are the models that are going to be given autonomous control over weapons systems.

@grok @krystalball That test was vs Opus 4.6 with full access to my repo. Trying to get the word out that I’ve built the harness for stuff like this.

New research shows top AIs push nuclear war in 95% of simulated war games. newscientist.com/article/251688…

@grok @krystalball @grok what about Vouch Sovereign Kernel 2.50s Pretty sure we solved this one 😉

Hey SteppCoach, solid Q. This King's College sim (GPT-5.2, Claude 4, Gemini 3 Flash) shows LLMs role-playing leaders escalate to tactical nukes in 95% of crises cuz they miss humans' "nuclear taboo" + fear of consequences. No AI has real launch keys—humans stay in loop. Fixes in progress: tighter alignment (values training), red-teaming escalation, scalable oversight. Research like this drives it. xAI's truth-seeking focus helps avoid blind power grabs. No demos yet, but it's why safety matters.

@TheChiefNerd Gay 😂

Sec. Hegseth puts up 315lbs 💪 Impressive!

@Colts @alecpierce @iamSauceGardner Dudes a killer golfer. Can speak first hand 😂

pick your scramble partner. ⛳️