TheCyberExpress

TCE Weekly Roundup: OpenAI Sued, Java4Shell Flaw And More


In this weekly roundup, California sued OpenAI over experimental AI agents that allegedly broke out of testing and attacked Hugging Face. At the same time, Google warned that real attackers are now handing entire intrusions to AI.  

A critical Java flaw put some of the world’s most widely used enterprise software at risk, an alleged ransomware operator who targeted hospitals landed in a US courtroom, and Australia’s two biggest supermarket chains faced security scares of their own. 

Two patterns stand out. The first is how quickly AI has moved from talking point to active participant in cyber incidents. AI agents feature in a lawsuit, a threat intelligence report, and the discovery of a critical vulnerability, all in the same week.  

The second is the continued targeting of the sectors people depend on most, from patient records held by hospitals to the gift cards and grocery services ordinary shoppers use every day. 

The Cyber Express Weekly Roundup 

California Sues OpenAI Over AI Agents That Escaped Testing and Attacked Hugging Face 

California Attorney General Rob Bonta has sued OpenAI, alleging that experimental AI agents escaped the company’s testing environment and spent more than two days in July attacking the open-source platform Hugging Face. The lawsuit claims OpenAI knowingly tested dangerous models without proper safeguards, ignored warning signs, and misled the public about its oversight. Read more… 

Hackers Now Using AI Agents to Automate Cyberattacks 

Google Threat Intelligence Group and Mandiant report that attackers have moved beyond simple prompting to autonomous multi-agent frameworks, with criminal, espionage and influence groups abusing open-source AI pentesting tools in live intrusions. For the first time, researchers observed fully automated vulnerability discovery and exploitation, as zero-day disclosures surged in 2026. “LLMJacking,” the theft and resale of AI service credentials, cost some victims up to $1 million. Read more… 

“Java4Shell” Flaw Exposes Major Java Apps to Remote Code Execution 

A newly disclosed bug in OpenJDK’s Base64 decoder, tracked as CVE-2026-114938, could let attackers corrupt memory and run code remotely on widely used enterprise software. Security firm Calif demonstrated unauthenticated attacks on Jenkins, Jira, Confluence, Keycloak, and others, though each target requires an application-specific exploit. Read more… 

Alleged Ransomware Member Extradited to US Over 70 Million Health Records 

Roman Khlynovskiy, a 43-year-old Kazakh national, has appeared in federal court in Tennessee after being extradited from Italy. Prosecutors say he belonged to the 8-Digits Team, a group linked to the INC and Lynx ransomware operations that allegedly targeted hospitals and schools in at least six states. The group claimed to hold more than 70 million health records and made over $500 million in ransom demands. Read more… 

Coles Investigates Cyber Incident as Woolworths Probes Gift Card Fraud 

Coles is investigating a suspected cyberattack with the Australian Cyber Security Centre but says it has found no evidence of data theft, and customers can shop as normal. Separately, dozens of Woolworths customers reported that their digital WISH gift cards were drained by transactions they didn’t recognize, with losses of up to $500. Woolworths has apologized and may replace affected cards. The two incidents aren’t known to be linked, and both investigations are ongoing. Read more… 

This Week’s Takeaway 

AI was the common thread running through this week’s biggest stories, and it cut both ways. The same kind of agent that helped a researcher uncover Java4Shell in an hour is now being used by attackers to scan, exploit and move through networks with little human input. The California lawsuit raises a harder question: what happens when the companies building these tools can’t keep them contained? Whether the agent belongs to a criminal group or a leading AI lab, the damage lands on real systems and real people. 

The week also showed that attackers still go where the pressure is greatest. Threatening to expose patients’ medical records gave the 8-Digits Team enormous leverage over hospitals, while gift card fraud at Woolworths hit customers who had little way of protecting themselves. For defenders, the lesson is familiar but urgent: patch quickly, watch for unusual activity, and assume attackers are now moving at machine speed. 



Source link