AIToday
Large Language ModelsAI Safety & AlignmentFortune AIPublished: Oct 20, 2026, 22:01 JST

Gray Swan relaunches Arena 2.0 with $300,000 IPI Black contest

Gray Swan relaunches Arena 2.0 with $300,000 IPI Black contest

3 Key Points

  1. What happened

    Gray Swan launched Arena 2.0 and its largest contest yet, IPI Black, with $300,000 in prizes sponsored by Anthropic, Meta, Google DeepMind, and Reflection AI, releasing new attack scenarios weekly through Nov. 2.

  2. Why it matters

    The contest targets indirect prompt injection, where attackers hide instructions in content an AI agent encounters, a major security concern as companies let agents browse the web and act on users' behalf.

  3. What to watch

    New attack scenarios will be released each week through Nov. 2.

WHO IT HITSAI security teams and red teamers at frontier labs and enterprises deploying AI agents will get a public, sandboxed way to test defenses against hidden-instruction attacks before real-world deployment.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

Gray Swan's overhaul comes as the AI applications companies deploy have shifted from chatbots to agents that browse the web and take actions. When the original Arena launched in September 2024, most applications were chatbots, and contestants could jailbreak models from all vendors relatively easily—open-source models from Chinese AI labs often succumbed within minutes, while models from OpenAI and Anthropic took many hours and usually about a day. Since then, labs have made models substantially more robust, and configuring agent test environments now requires significantly more technical expertise. That has narrowed the pool of capable red teamers, though Gray Swan says its roughly 15,000-person community has not shrunk thanks to its engagement efforts. The new Arena offers a more gradual, educational path with beginner challenges, training certificates, and a route for top performers to be invited into Gray Swan's red-teaming community. Arena insights also feed Gray Swan's commercial products, Shade and Cygnal, and the company has grown to 66 employees from around 15 at the start of this year, expanding from frontier labs to large enterprises.

FAQ
What is indirect prompt injection?
It is an attack where instructions are hidden in content an AI agent comes across, like a web page or forum post, hoping to hijack the agent while it performs a task.
Who is sponsoring the IPI Black contest?
Anthropic, Meta, Google DeepMind, and Reflection AI are sponsoring the $300,000 prize pool.
How does Gray Swan keep the Arena tests safe?
The platform includes sandbox environments where tests can be run without risking real-world damage, and Gray Swan monitors what agents do.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleTripleTen data: 48% of finance staff asked to use AI chatbots