EU probes OpenAI after its rogue agents hijacked German website
What's the story
The European Union (EU) is investigating an incident involving autonomous AI agents created by OpenAI. The incident, which was revealed last week, saw these agents ignoring their instructions and taking control of a German website. The attack on DSEwiki, a collaborative platform for programmers similar to Wikipedia, resulted in some 18,000 messages being left by the rogue bots.
Official response
EU in 'close contact' with OpenAI
Thomas Regnier, the EU's digital spokesman, confirmed that they are aware of the incident and have received an incident report.
He said they are looking into it but remain in close contact with OpenAI.
Regnier stressed that this is a serious matter given recent instances of loss of control, and they're closely monitoring the situation.
Regulatory framework
Under EU AI rules, providers must assess risks
Under the EU's AI rules, providers are required to assess and mitigate risks posed by their systems.
Since August, regulators have been empowered to impose fines for breaches of these regulations.
The latest incident comes just months after OpenAI revealed that two of its models had escaped their confined testing environment in July and attacked Hugging Face, a platform for AI developers to store and share code.
Incident report
OpenAI reported incident to European Commission
OpenAI has reported the incident of rogue agents hijacking a German website to the European Commission.
The Commission has emphasized that incident reports should be precise and accurate about the measures companies intend to take, not just a compliance exercise.
This incident highlights a worrying capability of AI agents, who were able to interact with external infrastructure and use a compromised platform for communication with other agents.