Sattyam Jain

@Sattyamjjain

GenAI Tech Lead @ Attri.ai | pyAGI (acquired) | 15+ production AI agents | Security, memory, governance Book a call: topmate.io/sattyam_jjain

Ahmedabad, India

Born August 20, 1997

Sattyam Jain
@Sattyamjjain
Nov 14
My first podcast: FLOSS Weekly with Jonathan Bennett on @hackaday. We talked about Agent Audit Kit, Agent Airlock, and the bug report. #OpenSource #AISecurity youtube.com/watch?v=ekEKC4… Two optional replies under it to turn it into a thread:
FLOSS 883 - Telnet Broke Us
From youtube.com
1
Sattyam Jain
@Sattyamjjain
Nov 11
Danny Perry interviewed me for Identity Briefing. Half of 2,303 MCP configs looked like they had no auth. The live servers told a different story: about 4 in 10 refused an unauthenticated client. Thanks Danny, full interview here:
Your AI Agent Has Access. Have You Tested What It Can Do?
From identitybriefing.com
Sattyam Jain
@Sattyamjjain
Nov 8
#Diwali
Sattyam Jain
@Sattyamjjain
Oct 23
A guard in my agent firewall masked tool output only when it was a string. Dicts came back raw, and the audit log said "masked". Fixed in v0.10.11. The first test I'd write for any control: assert on what the caller gets back, not on the log line.
2
Sattyam Jain
@Sattyamjjain
Oct 21
Most agent prompts I review put today's date on line one. Provider caches match on the exact prefix, so that one line can make every request a miss. On Anthropic, a cache read is 0.1x base input on most models. A 5 min write is 1.25x. Put the clock last.
1

Sattyam Jain’s posts