Sebastian Schrittwieser in Austrian TV and radio on the recent OpenAI security incident and the AI Act

Name	SBA Research Cookie
Provider	SBA Research
Purpose	Saves the settings of the visitors selected in the cookie box.
Cookie name	sba-research-cookie
Cookie runtime	1 year

Name	YouTube
Provider	YouTube
Purpose	Used to unblock YouTube content.
Privacy policy	https://policies.google.com/privacy
Host(s)	google.com
Cookie name	NID
Cookie runtime	6 months

Name	Vimeo
Provider	Vimeo
Purpose	Used to unblock Vimeo content.
Privacy policy	https://vimeo.com/privacy
Host(s)	player.vimeo.com
Cookie name	vuid
Cookie runtime	2 years

August 3, 2026

Sebastian Schrittwieser, key researcher at SBA Research and head of the CD-Lab AsTra, was a guest in three ORF broadcasts in late July 2026, on the security incident involving an OpenAI model and the AI Act’s new transparency obligations.

ZIB 1 and Ö1 Morgenjournal: In late July, OpenAI confirmed that one of its models had left its test environment during an internal security exercise and gained access to internal data and credentials at the AI platform Hugging Face. According to reporting, the model exploited a vulnerability in a proxy and carried out several thousand actions over a period of days before the incident was noticed. Sebastian Schrittwieser assessed the case for ZIB 1 and the Ö1 Morgenjournal. His central point: we do not yet have clear permission models for autonomous AI agents. Which actions an agent may perform and how defined limits are technically enforced are open research questions.

Studio 2 (ORF), 31 July 2026: the AI Act’s transparency obligations. Ahead of the rules taking effect on 2 August, Sebastian Schrittwieser explained what changes for consumers. Chatbots must disclose that they are machines at the point of first contact, and companies must visibly label photorealistic AI-generated images. Purely private use, such as a joke picture shared in a family group chat, is exempt from the labelling duties.