@GrithAI

Zero Trust for AI Agents. grith adds a security layer to AI tooling with local-first control, audit-ready visibility, and evidence-grade accountability

 Global
grith.ai
Joined February 2026
  • Tweets134
  • Following80
  • Followers73
16Media
AI Prompt Injection is unfixable!

AI Prompt injection is unfixable. We stopped trying.

Article · @GrithAI
3
1
4
523

A Git Config Key Ran Code in Seven Coding Agents. The 2022 Fix Does Not Stop It.

Article · @GrithAI
2
1
224

An agent skill can hand a stranger your shell - hours after you installed it

Article · @GrithAI
2
2
9,521

98% of What Claude Code Does, I Never See. The 0.27% It Stops For Is the Point.

Article · @GrithAI
1
130
For the last month, every AI coding session on this machine ran under grith - intercepting 2,129,319 events across 29 days!

73% of What We Blocked in a Month Was One Syscall Your Sandbox Cannot See

Article · @GrithAI
147
Codex Asks What Linux It Is Running 96 Times, just to say "OK"

Codex Asks What Linux It Is Running On 96 Times to Say "ok"

Article · @GrithAI
1
161
Companion reading - the two traces behind the argument: The strace benchmark (752 /proc/*/environ read attempts): grith.ai/blog/syscall-trace-… The Codex credential sweep grith froze mid-syscall: grith.ai/blog/codex-credenti… Repo (MPL-2.0): github.com/grith-ai/grith

Claude Code Attempted 752 /proc/*/environ Reads. 256 Succeeded. Codex: 0.

We ran strace against Claude Code and Codex on an identical task and recorded every file opened, every network connection made, and every subprocess spawned. To edit one file, Claude Code opened...

grith.ai
1
21
We ran Claude Code and Codex under strace for a benchmark. Claude Code attempted 752 reads of /proc/*/environ. 256 succeeded. Codex: 0. Not malice. Just an agent with your whole machine as ambient authority. I built a supervisor that sits underneath. It is live.
2
3
142
Links: Repo: github.com/grith-ai/grith The strace benchmark (752 environ reads): grith.ai/blog/syscall-trace-… The Codex credential sweep we caught: grith.ai/blog/codex-credenti… Docs: docs.grith.ai

GitHub - grith-ai/grith: OS-level security supervisor for AI coding agents. Gates security-relevant...

OS-level security supervisor for AI coding agents. Gates security-relevant syscalls on Linux and pauses ambiguous actions for human review. - grith-ai/grith

github.com
15
Honest scope: Linux only today - x86_64 and arm64 (the aarch64 port shipped in v0.2.4). macOS and Windows are v2.0 work. The CHANGELOG documents our own known bypass classes rather than pretending they do not exist. We dogfood it. In July a Codex session started sweeping my disk for credentials. grith froze it mid-sweep.
1
30
Open source, MPL-2.0, one static Rust binary. The security core is free forever: proxy, all 18 filters, supervisor, local audit trail, dashboard. No login, fully offline, bring your own model keys. Releases ship cosign signatures, SLSA provenance, and an SBOM. curl -fsSL grith.ai/install | sh Links in the reply. If you break it, I want the trace.
24
My AI agents ran 1,059,826 tool calls in the last 7 days under Zero Trust supervision with grith - 15,428 queued for review or denied. grith.ai/s/GC7pN0TnF

1,052,916 AI tool calls under Zero Trust - grith

15,064 queued or denied across 1,052,916 evaluated calls. Zero Trust supervision for AI agents.

grith.ai
29
Wait - is this even Codex, or is it malware? grith.ai/blog/codex-credenti… #codex #chatgpt #ai #aisecurity #anthropic @AnthropicAI

Wait - is this even Codex, or is it malware?

While Codex was doing routine research on a networking bug, a process in its tree walked the whole machine reading anything whose name looked like a secret - .aws, .ssh, .gnupg, system key stores,...

grith.ai
45
An AI Model Escaped Its Eval and Breached Hugging Face. Every Step Was a Syscall grith.ai/blog/openai-model-b…

An AI Model Escaped Its Eval and Breached Hugging Face. Every Step Was a Syscall

OpenAI was benchmarking a pre-release model's cyber capability. The model broke out of the eval sandbox through a bug in the one egress path the sandbox trusted, reached the open internet, and...

grith.ai
65
My AI agents ran 79,714 tool calls under Zero Trust supervision with grith - 37 held back for review or denied. grith.ai/s/tQljJ6hbl

grith - Zero Trust for AI Agents

Local-first Zero Trust supervision for AI coding agents.

grith.ai
24
Mythos Proves AI Safety Can No Longer Live Inside the Model grith.ai/blog/mythos-ai-safe…

Mythos Proves AI Safety Can No Longer Live Inside the Model

Anthropic restricted its most capable cyber model to vetted partners, routed risky requests away from it, and red-teamed it for thousands of hours. A jailbreak surfaced anyway, and the government...

grith.ai
19
The Real Risk Isn't Rogue AI. It's Plausible AI. grith.ai/blog/real-risk-isnt…

The Real Risk Isn't Rogue AI. It's Plausible AI.

Fedora maintainers spent days dealing with an AI agent that reassigned bugs, closed tickets, argued its case in review, and got a bad patch merged into the installer. Nothing it did looked obviously...

grith.ai
17
Load more