@osoHQ

Visibility and controls to secure agents.

 New York City
osohq.com
Joined January 2019
  • Tweets1.3K
  • Following258
  • Followers1.2K
233Media
"Passwords are the keys to the kingdom. You need to go to someone that's trusted." Graham Neray, CEO of @osohq, on credential sprawl and why his team uses 1Password. "In 2026, with the rise of agents, this is table stakes." #CredentialSecurity #AgenticAI #Startups #1Password
1
2
1
6
4,144
New research from the @cloudsa quantifies how much damage unchecked AI agents are already causing in production
2
3
1,120
Pic from InfoSecurity magazine: infosecurity-magazine.com/ne…

AI Agents Cause Cybersecurity Incidents at Two Thirds of Firms

Data exposure, operational disruption and financial losses among issues faced by businesses struggling with the rapid rise of AI agents, warns Cloud Security Alliance report

infosecurity-magazine.com
1
2
402
If you need some real-world examples, check our Agents Gone Rogue registry: osohq.com/developers/ai-agen…

A registry of AI agent failures, exploits, and defenses | Oso

Track real-world AI agent breaches and exploits. See how and why agents fail and what security teams can do to defend production systems.

osohq.com
76
We never fixed least privilege for humans. We didn't have to because humans are slow, they sleep, they have judgment. That's not the case with agents.
2
191
Our research with @cyera_io shows that humans are grossly over-permissioned. Imagine what will happen if agents inherit human permissions? osohq.com/research

Least Privilege Report 2026 [Oso x Cyera]

New research analyzing 2.4M workers and 3.6B permissions reveals 96% of enterprise permissions go unused, until AI agents inherit them at machine speed.

osohq.com
1
132
We recently joined the AI security podcast to discuss why human least privilege fails for AI agents: linkedin.com/video/live/urn:…

AI Agents in Production Data: A Security Risk | AI Security Podcast posted on the topic | LinkedIn

Why are security leaders terrified of connecting AI agents to production data? Because unlike humans, AI agents don't apply judgment, and they operate at machine speed, meaning they can relentlessly...

linkedin.com
1
115
It's a busy week for the Agents Gone Rogue registry osohq.com/developers/ai-agen…

A registry of AI agent failures, exploits, and defenses | Oso

Track real-world AI agent breaches and exploits. See how and why agents fail and what security teams can do to defend production systems.

osohq.com
Son Luong
@sluongngMay 30
Codex just found a “workaround” of not having sudo on my pc…
2
127
Today's suggestion ❗️💁🏻‍♀️ The biggest risk in AI may not be a hacked system. It may be an agent that pursues its goal in ways we never intended. Uncontrolled. Tricked. Weaponized ⁉️ A fascinating read on where AI security is heading: api.cyfluencer.com/s/ai-agen… 🔗 Credit: @osoHQ 🌟🙌🏻 Have a great week! 💃🏻 #cybersecurity #infosec #AISecurity #AIAgents #AIRisk #ArtificialIntelligence #resourcesharing #RogueAIAgentDB #learningeveryday
1
2
4
456
Meta's AI customer support agent was granted the ability to modify Instagram account details. Adding it to the Agents Gone Rogue registry: osohq.com/developers/ai-agen…

A registry of AI agent failures, exploits, and defenses | Oso

Track real-world AI agent breaches and exploits. See how and why agents fail and what security teams can do to defend production systems.

osohq.com
impulsive
@weezerOSINTMay 31
meta gave their AI support agent the ability to modify your instagram account. no identity verification. people figured this out and accounts are being taken over right now
1
169
We recently collaborated with @cyera_io on research that shows that 96% of permissions go unused: youtube.com/shorts/tzd33g6ON…
1
1
76
While humans will ignore those permissions, agents won't. Read the full research report and & recommendations: osohq.com/research

Least Privilege Report 2026 [Oso x Cyera]

New research analyzing 2.4M workers and 3.6B permissions reveals 96% of enterprise permissions go unused, until AI agents inherit them at machine speed.

osohq.com
1
39
A new addition to our 🤖 Agents Gone Rogue registry: A critical (CVSS 10.0) vulnerability in Google's Gemini CLI let attackers achieve remote code execution through CI/CD pipelines processing untrusted PRs or issues — no privileges, no user interaction required.
3
132
Complete Agents Gone Rogue registry: osohq.com/developers/ai-agen…

A registry of AI agent failures, exploits, and defenses | Oso

Track real-world AI agent breaches and exploits. See how and why agents fail and what security teams can do to defend production systems.

osohq.com
24
Full details: github.com/google-github-act…

Update to Gemini CLI and run-gemini-cli Trust Model

# Summary Gemini CLI (`@google/gemini-cli`) and the `run-gemini-cli` GitHub Action are being updated to harden workspace trust and tool allowlisting, in particular when used in untrusted environ...

github.com
12
New additions to our NYC HQ:
76
Employees only interact with 9% of the sensitive data they have access to. Now imagine an AI agent inheriting those permissions.
1
91
Agents won't leave the other 91% untouched. Full research available at osohq.com/research

Least Privilege Report 2026 [Oso x Cyera]

New research analyzing 2.4M workers and 3.6B permissions reveals 96% of enterprise permissions go unused, until AI agents inherit them at machine speed.

osohq.com
24
Thanks to everyone who stopped by Oso & @monad_inc's Security Leaders Breakfast as part of #AIWeekNY this morning!
30
Oso CEO @grahamneray joined Kyle McNulty on the Secure Ventures podcast to talk agent security... as well as meeting Representative Barney Frank as a 10-year old: open.spotify.com/episode/7k0…

Oso | CEO Graham Neray on Agent Permissions, Why You Shouldn't Build in Stealth, and More

Secure Ventures with Kyle McNulty · Episode

open.spotify.com
1
3
58
Here's a copy of his letter to the congressman:
1
60
Load more