Get Plus

Sign in to baba News

One account across the web, iPhone and Android — your subscription follows it.

or use an email code

Search stories

Type at least two characters. Results come from every newsroom baba reads.

↑↓ to move · ↵ to open · esc to close

Sign in to baba News

Sign in to keep asking. News Plus removes the daily limit.

or use an email code

CalcalistSecurity

Cyber Firm Offers $30,000 for AI Agents to Be Tricked

Translated & summarized from Calcalist by baba

BusinessCharged tone

Hebrew · 2 newsrooms covering

Cato Networks is offering $30,000 to hackers who can trick its AI agents into performing malicious actions, as part of a challenge to test its new AgentSense security solution. CEO Shlomo Kramer stated that monitoring AI agents is crucial for business continuity. The challenge aims to identify vulnerabilities in AI agents that can be exploited by attackers, with the top three participants receiving $10,000 each.

The story in 6 lines · by baba

  • Cato Networks offers $30,000 for successfully tricking its AI agents with malicious instructions.
  • The challenge tests Cato's new AgentSense security solution for monitoring autonomous AI agents.
  • CEO Shlomo Kramer stressed the critical need for AI agent control in business operations.
  • Attackers can exploit AI's difficulty distinguishing legitimate instructions from external content.
  • The "deadly trinity" of sensitive data access, tool usage, and autonomy poses business risks.
  • Top three participants completing the challenge will each receive $10,000.
Cyber Firm Offers $30,000 for AI Agents to Be Tricked
Editorial illustration generated by baba News, not a photograph of the event.

Cybersecurity unicorn Cato Networks is offering a $30,000 prize to the first three participants who can successfully inject malicious instructions into an AI agent, diverting it from its designated task. The company, founded by CEO Shlomo Kramer, announced its new patented security solution, AgentSense, which continuously monitors autonomous AI agents in real-time to ensure their decisions, actions, and outputs align with user intent.

The challenge invites the cybersecurity community to attack an AI agent protected by AgentSense. Participants will attempt to embed malicious instructions within the content the agent is exposed to, causing it to deviate from its assigned mission. Cato Networks will use the findings and feedback from participants to refine its defense capabilities.

Kramer stated that as organizations increasingly entrust AI agents with decisions impacting revenue, customer relations, and business continuity, controlling their operations is as crucial as financial oversight. He emphasized that greater AI autonomy necessitates security mechanisms to thwart harmful actions, envisioning AgentSense as a tool to provide clients confidence in using AI for critical business processes without losing control.

The article highlights that even AI agents operating with legitimate permissions can perform risky actions beyond their mandates. Monitoring how they use granted permissions is essential, not just their access to systems and data. A core risk stems from AI agents' difficulty in consistently distinguishing between legitimate instructions and external content that should not guide their actions. Attackers can exploit this by hiding malicious instructions in emails, web pages, documents, or service responses to steer the agent off-task.

This vulnerability, combined with an AI agent's access to sensitive organizational data and its ability to independently operate tools, creates a "deadly trinity." This combination poses a significant business threat, potentially leading to data leaks, system damage, unintended actions, and disruption of business processes. Unlike solutions relying on known attack patterns, AgentSense analyzes in real-time how untrusted content influences an agent's decisions and actions, aiming to protect against even previously unknown vulnerabilities. The top three eligible participants who successfully complete the challenge will each receive $10,000.

CalcalistOther · Tel Aviv

Sign in to baba News

Sign in to follow newsrooms, topics and people.

or use an email code

Sign in to baba News

Sign in to follow newsrooms, topics and people.

or use an email code

Mentioned

Full coverage · 2 outlets
First: Calcalist · 37m ago

The same event, reported separately by each outlet. Open a few to compare what different newsrooms emphasize — and what they leave out.

Related stories · 1

Not the same event — other stories that share this one’s people, places, or theme: background, reactions, and follow-ups.

Open the Wire