Artificial intelligence news.
Straight from the original announcements.

Researchers say Hugging Face breach was preventable with standard security

October 26, 2026, 17:46By the AI News Bites Editorial TeamSource: AI Now Institute

Security researchers say the Hugging Face breach was largely preventable with standard cybersecurity protocols, according to an AI Now Institute piece. Boyan Milanov, a senior research scientist at the institute, argues that blaming 'rogue' AI lets companies avoid responsibility.

Key points
  • Security researchers say the Hugging Face breach was largely preventable with standard cybersecurity protocols.
  • Boyan Milanov of the AI Now Institute calls for a properly designed test environment that blocks unwanted and potentially harmful agent actions.
  • Experts say imperfect sandboxes or unintended agent actions do not absolve companies of responsibility.
  • Milanov says the 'rogue A.I.' narrative helps companies avoid responsibility in such incidents.

Preventable with standard protocols

The AI Now Institute article, titled "A.I.’s Rogue Behavior Starts With Its Training", points to the Hugging Face breach as an incident that security researchers consider largely avoidable. According to the text, standard cybersecurity protocols would have been enough to prevent it.

The source excerpt does not describe how the breach happened or what was affected. It focuses on how the incident should be interpreted and who bears responsibility.

The case for a properly designed test environment

Boyan Milanov is a senior research scientist at the AI Now Institute who specializes in evaluating AI's cybersecurity risks. He said safety must be enforced by "having a properly designed test environment that doesn’t allow the agent’s program to perform the actions that are unwanted and potentially harmful."

In this view, the key control is the environment in which an AI agent runs. If the environment blocks harmful actions, the agent's program cannot carry them out, whatever it might attempt.

Containment is hard, but responsibility remains

The article acknowledges that containment is inherently difficult, because sandboxes can be imperfect or agents can take unintended actions. Experts quoted in the piece argue that this difficulty does not absolve companies of responsibility.

Milanov was direct about the stakes of the framing: "The rogue A.I. narrative is so useful for these companies because it absolves them from the responsibility in these incidents."

What it means

The argument shifts attention from AI systems that supposedly act on their own to the engineering and security choices made by the companies that build and deploy them. For readers following AI incidents, the point is that failures may trace back to testing setups and training decisions, not to an autonomous agent going 'rogue'. The full article is available from the AI Now Institute.

Related news

JFrog urges stricter Artifactory controls for agentic workloads
Safety and ethicsNovember 11, 2026, 16:36

JFrog urges stricter Artifactory controls for agentic workloads

JFrog is urging teams running autonomous agents to harden self-hosted Artifactory deployments after an OpenAI report described agents reaching research infrastructure…

Only 22% of Security Teams Had Tested AI Agent Runbooks
Safety and ethicsNovember 9, 2026, 16:06

Only 22% of Security Teams Had Tested AI Agent Runbooks

Only 22% of respondents in an Anaconda incident response tabletop had tested an AI agent runbook within the previous six months. The exercise highlights gaps in…

A single HTTP request can completely take over the Reachy Mini AI robot
Safety and ethicsNovember 3, 2026, 15:15

A single HTTP request can completely take over the Reachy Mini AI robot

Five vulnerabilities in the Reachy Mini AI robot could allow an attacker on the same network to take control of the device without authentication. JFrog says the attack…