Our articles, videos, and infographics inform our readers about developments in technology, engineering, and science. A public charity, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity. Access to Spectrum's Digital Edition is exclusive for IEEE Members Create an account to access more content and features on IEEE Spectrum, including the ability to save articles to read later, download Spectrum Collections, and participate in conversations with readers and editors. For more exclusive content and features, consider Joining IEEE . Join the world's largest professional organization devoted to engineering and applied sciences and get access to all of Spectrum's articles, archives, PDF downloads, and other benefits.

Today, most of us interact with AI assistants - reactive bots that wait for human instructions. Yet, AI assistants are rapidly being replaced by agentic AI agents, that can interact with external tools, browse the web, generate images, send emails, and perform increasingly complex workflows on behalf of users. As these agentic agents become more capable, they may be exploited by people with bad intentions. Safety tests have generally only checked whether an AI agent refuses a single harmful request - they don't measure what happens when an attacker gradually persuades the AI through a series of seemingly harmless conversations. Now, researchers from EPFL's Natural Language Processing Laboratory have developed STING (Sequential Testing of Illicit N-step Goal execution), an automated testing framework that simulates how a real attacker might manipulate large language model (LLM) agents into carrying out harmful tasks over multiple interactions.

Say More Look Up Mashable's Best: E-readers, robovacs, laptops, earbuds, smart home and more Mashable Selects Trending Now Gift Ideas For Everyone On Your List Safety Net Mashable @Home In My Bag Versus Creator Playbook Getting Technical with Mashable All Series AI agents may run your household. Some parents are eager for AI agents to reduce their mental load, but the privacy risks are real. Rebecca Ruiz is a Senior Reporter at Mashable. She frequently covers mental health, digital culture, and technology. Her areas of expertise include suicide prevention, screen use and mental health, parenting, youth well-being, and meditation and mindfulness.

Perplexity has launched its personal AI agents called teammates, which are designed to be able to tackle ongoing work autonomously with minimal supervision. The agents work inside Perplexity Computer, the company's cloud-based agentic assistant that can perform multi-step tasks for you. Perplexity explains that the new agents build on top of Computer and come with additional capabilities for ongoing, large-scale work. You'll be able to assign each teammate its own role or task that they can all then perform simultaneously. Teammates work together, and they can even manage other teammates to make sure your tasks are getting done.

Mashable Selects Mashable's Best: E-readers, robovacs, laptops, earbuds, smart home and more Look Up Say More Trending Now Gift Ideas For Everyone On Your List Safety Net Mashable @Home In My Bag Versus Creator Playbook Getting Technical with Mashable All Series If bots message for daters, there's a low chance they'll actually connect, Match Group figures. Anna Iovine is the associate editor of features at Mashable. Previously, as the sex and relationships reporter, she covered topics ranging from dating apps to pelvic pain. Before Mashable, Anna was a social editor at VICE and freelanced for publications such as Slate and the Columbia Journalism Review. Follow her on Instagram .

The Dell Laptop Buying Guide 2026: Which XPS or Alienware Laptop Should You Buy? The Dell Laptop Buying Guide UK: Which Dell Laptop Should You Buy? When you purchase through links in our articles, we may earn a small commission. My AI agents sent emails without my OK. AI agents tend to act first without permission.

The startup Aina bets you won't need to rely on your phone to control virtual assistants. It's making a smart ring that lets you control the actions of your autonomous agents with one tap. Apoorv Shankar believes that in the very near future, humans will interact with technology through a series of approvals and rejections. AI agents will proactively come to us to complete tasks, and humans will say, "Yes, order that Uber," or "No, don't get lunch from that spot." But if you get multiple pings from an AI agent throughout the day, taking out your phone to say yes or no every single time will be a "bottleneck," he says.

Data lineage is the new security perimeter, providing the visibility needed for governance, accountability, and trust, without impeding AI adoption at scale. A sales representative asks an AI agent for customer pricing, and the agent pulls data from multiple sources: restricted pricing systems, internal communications, and archived quotes. The output looks accurate to the agent, so a second agent uses the data to draft the proposal for the sales rep. When it's discovered that the proposal includes unapproved, confidential, and sensitive information, no one can trace where the problematic data entered the workflow or how it spread across systems, files, and agents. This scenario illustrates why AI governance needs to evolve, rapidly and dynamically. As more enterprises begin to use agentic AI in daily workflows, they're realizing that they must manage more than static files; they must manage moving and fragmented data that is continuously ingested, copied, summarized, transformed, reused, and redistributed in new forms.

Plus: Today's midterms will test AI's influence on US politics. How "reverse game theory" could keep AI agents in check A new approach known as "reverse game theory" aims to control powerful AI models by designing systems that incentivize them to stick to the rules. While game theory predicts how rational actors behave under particular rules, reverse game theory designs rules to elicit the behavior we want. Think of splitting a cake: if one person cuts it and the other gets first choice, the cutter has an incentive to make the portions equal. Researchers believe the approach could align AI more closely with our rules and instructions. But big questions remain about whether these mechanisms will work in practice.

Amazon is blocking rival shopping agents, and even websites that welcome AI assistants can stop them with an ordinary CAPTCHA. An AI agent can take over your grocery shopping, right up until a website asks it to prove it's human. That's the problem Meta's Muse has encountered at Walmart (one of its retail partners), which says the verification checks weren't intended to stop the assistant. Amazon has been more direct, deliberately blocking the tool from shopping on its site, according to TechCrunch . The appeal of personal AI agents like Meta's Muse is that these assistants can navigate websites, fill shopping carts and book travel instead of handing you recommendations and leaving the rest to you.