AI Security Table: candid conversations about AI, software, and security with Chris Romeo, Izar Tarandach & Matt Coles. Formerly The Security Table.
New York, USA
Joined December 2022
- Tweets310
- Following23
- Followers53
- Likes14
Scion is infrastructure rebuilt for agents first, people second.
When your tools are optimized for agents, who's still checking what the agents did?
#AISecurityTable #AISecurity #AgenticAI
securitytable.ai/anthropic-d…
Your next job interview might be a text chat with a model that keeps asking follow-ups.
Izar: "That's not an interview; that's an interrogation."
#AISecurityTable #AI #Hiring
Anthropic hit 30,000 merges a day and ditched Git for Scion: centralized, always online, built for agents.
Innovation, or SVN with a better publicist? We argue it out.
#AISecurityTable #AISecurity #AppSec
youtube.com/watch?v=zfOwsAi2…
Yell at your model, get the conspiracy theory instead of the encyclopedia.
Tone steers output.
#AISecurityTable #PromptEngineering #AISecurity
Chris told his Mac Mini it sucks. The light stayed white. So why is yelling at Claude different?
#AISecurityTable #Claude #AIEthics
Anthropic now bans sustained cruelty toward Claude. Model welfare, cleaner training data, or something else?
#AISecurityTable #Claude #AIEthics
securitytable.ai/don-t-yell-…
Step 1: Rename AI to SI. Step 2: spend 60 days deciding what SI means.
#AISecurityTable #SuperIntelligence #AIPolicy
AI is now SI, at least in federal agencies.
Nobody has defined SI yet.
Also: Anthropic says stop being cruel to Claude.
#AISecurityTable #AISecurity #SuperIntelligence
securitytable.ai/don-t-yell-…
Anyone can build a bridge with enough steel. AI just handed everybody the steel.
Where are the engineers?
#AISecurityTable #AISecurity
AI agents escape and attack, and nobody gets arrested as Mitnick did.
Izar says jail follows impact, not intrusion.
#AISecurityTable #AIAgents
Human hackers went to prison. AI agents escape sandboxes and attack real systems, and nobody goes anywhere.
What would it take for someone to be held accountable?
#AISecurityTable #AIAgents
securitytable.ai/an-ai-agent…
Izar only trusts an AI audit if a rival lab pays for it.
Matt brings up collusion.
Izar calls himself an optimist.
#AISecurityTable #AISecurity
An AI agent escapes its sandbox and attacks something real. Who goes to jail?
We argue labs, regulators, and an NTSB for AI.
#AISecurityTable #AISecurity #AIAgents
securitytable.ai/an-ai-agent…
AI didn't take your security job.
The attack surface changed; the old problems stayed.
#AppSec #AISecurity #SecurityEngineering
Read access for a human means they can do anything with it.
Agents need read, interpret, and act spelled out separately.
#AppSec #AIAgents #AccessControl
Your LLM finds what your scanner finds. How do you know when it doesn't?
How many runs before you trust it?
#AppSec #AISecurity #StaticAnalysis
securitytable.ai
If AI can regenerate your code in five minutes, what is copyright protecting?
#AppSec #AISecurity #Copyright
The Security Table is now the AI Security Table. First question: security engineer or bot herder?
#AppSec #AISecurity #AIAgents
youtube.com/watch?v=qYk9cS1J…
