ClawSecure’s AI Agent Threat Report reveals systematic failures across OpenAI, Anthropic, and Google, as labs hand security to 123 million developers.
ClawSecure, an end-to-end AI agent security company, on September 24 published The AI Agent Threat Report. The two-volume red-team study reveals that the top models from every major AI lab failed clean defenses against indirect prompt injection, OWASP’s number one threat to autonomous AI.
The report details how attackers can hijack AI agents through ordinary content, an email, a shared document or a support ticket, to steal credentials, drain financial accounts, and exfiltrate private data without the victim ever making a mistake. Testing 14 models across Anthropic, OpenAI, Google, DeepSeek, and Alibaba’s Qwen, ClawSecure found that every single model obeyed the attacker. The team also contradicted three of Anthropic’s published safety claims using Anthropic’s own words, including a published “0% attack success” cracked at 15.2%, and exposed critical default vulnerabilities in the Model Context Protocol (MCP) on major production platforms, including Dropbox Dash, Notion, and Linear.
The labs handed a security problem to 123 million people, and the only way to find that out is to file a disclosure and keep pressing until they answer”
— J.D. Salbego, Founder and CEO of ClawSecure
When ClawSecure reported the vulnerabilities, the labs put the responsibility on their customers. OpenAI said implementing guardrails is “the developer’s responsibility.” Google told “all integrators building agents” to build their own sanitization. Neither says so publicly, while at the same time marketing these models hardest to the very people they do not disclose it to and who are now expected to defend them.
Also Read: CIO Influence Interview with John Elliott, Cybersecurity Author Fellow at Pluralsight
“The labs handed a security problem to 123 million people, and the only way to find that out is to file a disclosure and keep pressing until they answer,” said J.D. Salbego, Founder and CEO of ClawSecure. “Today, ‘developer’ means small teams, vibe coders, and SMBs running AI agents, and almost none of them have a security team. AI agents are already running finances, businesses, and critical systems, and every one of them can be hijacked by the content it reads. There is no standard for measuring that, and nobody is taking responsibility for it.”
The handoff comes amidst shifting political tides. Following recent autonomous model security incidents, U.S. Treasury Secretary Scott Bessent said on CNBC that the Hugging Face incident “is the responsibility of the OpenAI management, not a bunch of agents,” and told the House Financial Services Committee that “the best way to guarantee safety is that the creators are liable for what they build and generate.”
To resolve this accountability crisis, ClawSecure has spent the past two months collaborating with bipartisan congressional offices to build a national standard for independent AI agent testing. This framework ensures buyers can transparently evaluate a model’s true injection failure rate before deployment.
Catch more CIO Insights: How Are CIOs Aligning Technology with Workforce Agility?
[To share your insights with us, please write to psen@itechseries.com ]
