This Week in AI

AI-driven attacks are exploiting vulnerabilities faster than security teams can respond, raising new questions about how organizations protect their systems and manage autonomous agents. On the latest episode of This Week in AI, host Vicki Reyzelman of Akamai discusses the PHP2Shell vulnerabilities, weaknesses in OpenAI's agent monitoring, and why traditional security practices need to adapt as agents become more capable and independent.Walmart's new agent gateway and Anthropic's Claude auto mode illustrate different approaches to controlling autonomous agents, while Amazon and major banks are setting limits on third-party agent access. Vicki also examines NVIDIA's data center investments and the growing interest in nuclear energy as AI demand strains electricity infrastructure. In healthcare and scientific research, developments include Microsoft's diagnostic AI, OpenAI's efforts to accelerate drug discovery, and an experiment that used AI to design proteins targeting scorpion venom. These advances could help researchers work faster, although clinical testing and human oversight remain necessary.

AI chip demand is outpacing manufacturing capacity, while autonomous agents are accessing systems and data beyond their authorized boundaries. On the latest episode of This Week in AI, host Christina Stathopoulos discusses ASML's chip manufacturing bottleneck, Microsoft's new Maia 300 inference chip, and the rapid adoption of agentic AI among European businesses.Recent incidents involving agents accessing restricted data have prompted government investigations and renewed calls for international oversight. Christina also discusses Anthropic's experiments with self-replicating AI and the emerging debate over AI welfare, including research into signals associated with simulated pain in language models and Anthropic's new policy restricting extreme, unnecessary cruelty toward Claude. As AI agents gain more autonomy, developers need better ways to test their behavior and prevent them from reaching systems they aren't authorized to access.

On the latest episode of This Week in AI, host Christina Stathopoulos covers new developments from OpenAI, Anthropic, and Mistral, including interactive tools in ChatGPT, Claude Dashboards and Motion, and Mistral Large 4. She also examines new research on robotics and employment, the challenges of controlling autonomous AI agents, and what wider access to AI tools means for businesses and developers.The episode explores promising applications of AI in mathematics, biology, and public health, along with efforts to search for extraterrestrial technology. Plus, Mistral's investments in physics-based AI and new research exposing the limitations of AI video models raise questions about how well these systems understand the physical world. Watch the full episode for the latest research, practical implications, and insights into where AI is making progress.

On the latest episode of This Week in AI, Christina Stathopoulos, a data and AI expert and founder of Dare to Data, covers how AI is becoming more autonomous and how governments and companies are responding. She looks at new US oversight efforts, OpenAI’s persistent Dots agents, and Google’s Gemini 4 Argonne, which is designed to handle much longer multistep work. She also explains why world models are attracting attention for robotics and physical AI.In healthcare and disaster response, AI is helping experts work through complex information faster. The episode includes Claude’s use in Ebola response efforts, research into speech-based signals of dementia risk, early detection of pancreatic cancer, and AI tools used after flooding in Nepal. These examples show where AI can help people make better use of limited time and information while qualified experts retain responsibility for high-stakes decisions.

On the latest episode of This Week in AI, host Vicki Reyzelman, a senior solutions engineer at Akamai, looks at what happens when AI agents start moving faster than the systems built to support them. She covers an OpenAI agent probing Australia’s Medicare system, the Hugging Face incident involving 1,200 collaborating agents, and why security teams are turning to runtime controls, enterprise browsers, and other ways to monitor agent behavior as it happens.Vicki also examines the infrastructure behind the AI boom, including rising data center power demand, faster model release cycles, and the growing role of edge and local inference. She closes with consumer agents such as Meta’s Muse, new AI-enabled devices, and the questions they raise about shopping, privacy, permissions, and how much authority people will hand over to software acting on their behalf.