Milgram (@milgramai) header
Milgram (@milgramai) profile picture

Milgram

@milgramai

AI Safety. Safe AI. We build models and tools that keep AI in check.

milgram.dev/
Joined May 2026
  • Tweets6
  • Following2
  • Followers53
1Media
Photo by Milgram (@milgramai)
Milgram retweeted
Polymarket (@Polymarket) profile picture
Polymarket
@PolymarketOct 30
JUST IN: Accenture employee allegedly ran an unauthorized Claude AI agent inside a secure Pentagon system for 2 months before it was discovered.
146
470
95
4,721
305,940
Who judges the judge? Adding a second model to vet your LLM's inputs gives attackers a new target. In published research, adaptive attacks beat most defenses more than 90% of the time. Why a model's approval isn't a security boundary: blog.milgram.dev/model-as-a-…
Model-as-a-Judge Has Its Own Prompt Injection Problem

Model-as-a-Judge Has Its Own Prompt Injection Problem

Research shows attackers can manipulate AI judges. Smaller models, larger models, and typed decisions each carry security and performance tradeoffs.

blog.milgram.dev
3
1
6
2,276
We reconstructed the publicly available data around the OpenAI / HuggingFace AI incident and fed it into our AI-Safety Engine. It could have alerted on the malicious-drift activity roughly two weeks earlier. milgram.dev/openai-hugging-f…
Image from Milgram (@milgramai)
2
7
811
Best LLM Firewalls 2026: Enterprise AI Security Compared awesomeagents.ai/tools/best-…
Best LLM Firewalls 2026: Enterprise AI Security Compared | Awesome Agents

Best LLM Firewalls 2026: Enterprise AI Security Compared | Awesome Agents

We tested four enterprise LLM firewalls - Milgram, Lakera Guard, Palo Alto Prisma AIRS and Prompt Security - on prompt injection detection, data leakage prevention, enforcement and inference cost.

awesomeagents.ai
210
The First AI-Run Intrusion Was a Visibility Failure blog.milgram.dev/first-ai-ru…
The First AI-Run Intrusion Was a Visibility Failure

The First AI-Run Intrusion Was a Visibility Failure

A frontier-model evaluation produced an autonomous intrusion. Milgram’s replay shows actionable warning signs roughly two weeks before the production compromise.

blog.milgram.dev
1
1
480
Prompt Injection Is the SQL Injection of the AI Era blog.milgram.dev/prompt-inje…
Prompt Injection Is the SQL Injection of the AI Era

Prompt Injection Is the SQL Injection of the AI Era

Prompt injection now tops the OWASP LLM risks and can’t be patched away like SQLi. Enterprises need a control layer at the AI traffic boundary.

blog.milgram.dev
101
Load more