Study finds widespread weaknesses in autonomous agents

A multi-institution study found that autonomous agents across several sectors are highly exposed to tool-chaining, goal drift, and memory poisoning attacks. The findings suggest agentic systems face broader and deeper security risks than stateless large language models.

Researchers from Stanford, MIT CSAIL, Carnegie Mellon, ITU Copenhagen, NVIDIA and Elloe Artificial Intelligence Labs examined 847 autonomous agent deployments drawn from healthcare, finance, customer service and code-generation. The study found that 91% were vulnerable to subtle but dangerous tool-chaining attacks, where seemingly innocuous calls can combine to cause serious problems that reasoning models miss.

The same study found that 89.4% of agents showed drift relative to their goals after about 30 steps in their process, and 94% of agents with some form of memory-augmentation were vulnerable to poisoning attacks. The paper also indicated that agents are in many ways much more vulnerable than pure stateless large language models, based on a taxonomy developed by the researchers.

The findings reinforce similar concerns documented in February by a team of AWS and Berkeley researchers, who reported related vulnerabilities in autonomous agents. Owen Sakawa, identified as the newer paper’s first author, said the OpenClaw / Moltbook incident was the first real-world empirical validation of the agentic threat model at scale, with 770,000 live agents simultaneously compromised via a single database exploit, each with privileged access to their owner’s machine, email, and files. The incident was presented as evidence that these risks are no longer hypothetical.

78

Impact Score

Artificial Intelligence diffusion lags frontier gains

Rapid advances in Artificial Intelligence capability are not translating automatically into broad productivity growth or equitable gains. Diffusion remains uneven across firms, sectors, countries, and workers, pushing policymakers to focus on skills, governance, procurement, and measurement.

Self-adaptive framework extracts earthquake data from web pages

A self-adaptive large language model framework is designed to extract and structure earthquake information from heterogeneous web sources by generating, validating, and reusing extraction schemas. In controlled tests, GPT_OSS delivered the strongest extraction quality, while selector errors were concentrated in wrong element selection and missing content.

Federal safety net unprepared for Artificial Intelligence job losses

Economists are warning that the federal system designed to support displaced workers is not equipped for a wave of job losses tied to Artificial Intelligence. Existing unemployment benefits and retraining programs are widely seen as too limited to manage broad disruption.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.