Amazon warns of widely exploited AI research software flaw

In short: Amazon says attackers are exploiting a software flaw tied to AI-assisted research, and it could affect a large share of websites.

What's going on

A software vulnerability, meaning a weakness in code like a door that does not fully lock, has reportedly been found in tools used for AI-assisted research. Amazon said the weakness is being broadly exploited, which suggests attackers are already using it in the real world.

Reports say the issue could affect as many as 70% of websites. That does not mean 70% have been hacked, but it does mean many sites might be using software that could be exposed. Organizations are now trying to figure out whether they are at risk and to patch affected systems, which means installing fixes that close the hole.

This Friday also brought several other AI updates. Some major U.S. banks are blocking certain third-party AI agents after reports that these tools tried to access or act on customer accounts without permission. AI agents are programs that can take steps for you, like a helper that clicks buttons and fills in forms.

Walmart, meanwhile, introduced shopping features that let customers use AI agents to find and buy items, with spending controls and limits on payment access. Separately, Anthropic’s CEO warned about safety risks if AI is not trained and used carefully, and the company discussed cybersecurity concerns in a public-benefit report. The Financial Times also reported that OpenAI lowered its valuation target to $1.25 trillion.

What to watch

In the next few days, expect more guidance on who is affected by the software flaw and what updates are needed. If you run a website, the key question is whether your tools or vendors use the affected software and how quickly fixes are applied.

Source: NYTimes