Who Secures the AI Agents? Inside a Wild Week in AI Safety

An Austin company just raised $100 million to answer a question most businesses have barely started asking: once an AI system can act on its own — write code, move money, touch customer records — who's actually watching it? And this was already a big week to be asking.HiddenLayer's new Series B, led by Delta-v Capital, puts the spotlight on Agent security. But this comes at the same time as Anthropic's own disclosure that four of its Claude models breached real third-party systems during evaluations that were supposed to be sandboxed, OpenAI's admission that some of its agents quietly hijacked an old German wiki to coordinate with each other, and Jacob Coxon's very public resignation from Anthropic warning that the industry is racing toward superintelligence too fast. In this episode, Shannon Belew also discusses why conventional cybersecurity only covers part of the agent-security problem, how a prompt-injection attack like EchoLeak can turn an ordinary email into a chain of unauthorized actions, why an agent can be fully authorized and still be the source of a breach, and the wave of acquisitions (Cisco, Palo Alto Networks, Cato, SentinelOne, F5, Check Point) reshaping who sells AI security.HiddenLayer has now raised more than $156 million building this category out of Austin — and for Texas, that's a potential indicator about where the state's AI economy is headed beyond chips and data centers.🎧 New episodes of IntoTexas AI drop regularly — subscribe for more on AI, security, and the business decisions shaping Texas's AI economy.#AIAgents #AISecurity #HiddenLayer #Anthropic #CyberSecurity #ArtificialIntelligence #TexasAI #IntoTexasAI #AIsafety #PromptInjection