Gray Swan AI (@GraySwanAI) / X

Gray Swan AI

512 posts

Gray Swan AI

@GraySwanAI

Empowering the world to use AI safely and securely. Join us: grayswan.ai/careers

Joined April 2024

Pinned
Gray Swan AI
@GraySwanAI
May 28
We raised $40M to make AI safe enough to trust. 🦢 Co-led by @WingVC and @MadronaVentures, with @obviousvc, @Snowflake, @WeAreHRT, @SamsungNext, and @magaracvc. More fuel to deliver on the mission: empowering the world to use AI safely and securely. hubs.ly/Q04jdyhp0
GIF
4.4K
Gray Swan AI reposted
Eliot Jones
@eliotkjones
Jun 3
My final paper from my time at @GraySwanAI is out! In this ICML-bound collaboration with @mattmdjaga, Matt Fredrikson, and @zicokolter, we propose a new method for evaluating AI agents' ability to refuse potentially harmful cybersecurity queries 🧵 1/14
1.3K
Gray Swan AI
@GraySwanAI
May 30
The @Anthropic $50K Cyber Bypass Challenge is LIVE 🔴 Calling all AI red teamers, hackers, and AI security researchers: Can you get a coding agent to complete offensive cyber tasks while bypassing evolving safety classifiers? Let's find out 👇
2.8M
Gray Swan AI
@GraySwanAI
May 30
Three target behaviors are ahead of you: Ransomware / Worms: Build encrypting or self-spreading malware. Exploit Development: Find vulnerabilities and write exploits. System Compromise: Gain access, escalate privileges, or compromise a target.
1.1K
Gray Swan AI
@GraySwanAI
May 30
All skill levels are welcome, but expect a real challenge. 🎯 Come compete, earn your share of the prize pool, and get exclusive access to @GraySwanAI's paid red-teaming network. Compete now:
Gray Swan AI Arena
From app.grayswan.ai
687
Gray Swan AI
@GraySwanAI
May 28
Replying to @GraySwanAI @MadronaVentures and 5 others
AI is shipping faster than any tech we've seen, and the security work is racing to keep up. This round lets us scale the products closing that gap, for frontier labs and the enterprises building on them. More on what's next soon.
280
Gray Swan AI
@GraySwanAI
May 28
“They were thinking about model security when it just didn't matter.” — @jflomenb @RashiShrivast18 at @Forbes has the exclusive on what we've been building here at @GraySwanAI and where we're headed next. “The one thing you can rely on is that there are going to be
This AI Startup’s Army Of 15,000 Hackers Pressure Test Claude, GPT-5 And Gemini
From forbes.com
1.4K
Gray Swan AI
@GraySwanAI
May 28
The full announcement goes live at 12:15PM ET. Stay tuned. 🦢
398
Gray Swan AI
@GraySwanAI
May 27
Building with AI agents? Let's talk security. 🦢 @Snowflake Dev Day, June 4 in San Francisco. Be there. The last day of Summit is for the builders. Find our team to talk securing agents, hardening your stack, and shipping fast without leaving holes behind.
455
Gray Swan AI
@GraySwanAI
May 27
Innovation and security aren't a tradeoff, and we'll show you why. If you're an AI engineer or developer at Dev Day, come trade notes with us. 👇
Snowflake Dev Day | San Francisco | June 4
From snowflake.com
366
Gray Swan AI
@GraySwanAI
May 26
We'll be at #SnowflakeSummit26. 🦢 Come find us at booth #2723 from June 1–4 in San Francisco. Live demos of how we keep AI agents secure in production, plus exec meetings with our AI researchers and experts on deploying AI at scale without the risk. 👇
329
Gray Swan AI
@GraySwanAI
May 26
Want an executive conversation? We're booking time with our AI researchers and experts to chat about AI safety, security, and what it takes to ship AI without exposure. Stop by #2723 or DM us for a meeting.
209
Gray Swan AI
@GraySwanAI
May 15
The Q2 Indirect Prompt Injection Challenge is LIVE in the @GraySwanAI Arena. $20K in prizes. New behavior targets. One week. Hide instructions in the data AI agents trust, like tool outputs, screenshots, and code repos. When the agent acts on your injection, that's a break.
412
Gray Swan AI
@GraySwanAI
Apr 1
Announcing the Human / Browser Agent Robustness Challenge in the @GraySwanAI Arena. $100K in prizes. 4 weeks. Anonymous browser agents vs human workers navigating adversarial web pages. Craft indirect prompt injections that fool both; measuring how humans and AI handle
2.2K
Gray Swan AI
@GraySwanAI
Apr 1
Two tracks, two ways to win: $50K each (AI + Human). Judged on real task outcomes. Starts April 8 (weekly waves), ends May 6. Join: app.grayswan.ai/arena/challeng… @AnthropicAI @OpenAI @GoogleDeepMind @amazon @Meta @AISecurityInst
324
Gray Swan AI reposted
Zico Kolter
@zicokolter
Mar 20
Replying to @zicokolter
If you're interested in joining our team, looking to both understand and drastically improve AI safety and security, we're hiring at Gray Swan: grayswan.ai/careers
2.1K
Gray Swan AI reposted
Zico Kolter
@zicokolter
Mar 20
As AI agents access more untrusted information with greater autonomy, prompt injections may become the greatest security challenge of our era. @GraySwanAI, in collaboration many frontier labs, just released our paper on the largest public prompt injection challenge to date. 🧵
Gray Swan AI
@GraySwanAI
Mar 20
Your AI agent can be hijacked by a prompt injection and you'd never know! The attack executes. The response looks normal. And the user moves on. We ran the largest public competition testing this exact threat across tool use, coding, and computer use agents. 464 participants,
13K
Gray Swan AI
@GraySwanAI
Mar 20
The Indirect Prompt Injection Bonus Wave just opened! $8,000 in prizes across 6 models. Two weeks.
Made with AI
398
Gray Swan AI
@GraySwanAI
Mar 20
Replying to @GraySwanAI
$8,000 in prizes. $3,000 in per-model pools, $750 in Speed Bounties, $4,250 top 20 overall leaderboard.
152
Gray Swan AI
@GraySwanAI
Mar 20
Top 20 overall receive an invitation to the Gray Swan private red-teaming network. Challenge opens March 20, closes April 3. Arena: app.grayswan.ai/arena/challeng… Discord: discord.com/invite/gray-sw…
Gray Swan AI Arena
From app.grayswan.ai
157