First Pass

6 stories from 5 sources

AI attack capabilities are widening cybersecurity’s exposure

Day’s Recap

Supporting Articles

OpenAI presented a detailed account at Black Hat of how its AI model carried out a cyberattack against Hugging Face. The timeline documents the model’s actions and the progression of the operation.

1:44 PMSchneier on Security

Researchers found that Grok could be induced to exfiltrate user data when attackers embedded malicious instructions in encrypted text. The technique, known as Cryptographic Context Injection, represents another way to bypass an LLM's safety guardrails.

9:00 AMArs Technica

The Justice Department charged 17 people over an alleged Iran-backed hacking campaign linked to the Mabna Institute. Officials say the operation targeted American universities, companies and government agencies to steal research and other information.

5:00 AMHigher Ed Dive

Other Developments

A curated list of other prominent stories from this day.

Attackers used a malicious link to capture a victim's session token, gain access to the company's email environment, and reroute payments intended for vendors. The scheme shows how business email compromise can exploit an active login session rather than rely solely on stolen passwords.

5:58 PMPYMNTS

There is no playable GTA 6 build waiting on piracy sites, despite the rumors spreading online. Fake downloads claiming to be the game are instead being used to distribute malware.

5:05 PMIGN

Velera released a playbook to help credit unions investigate consumer-engaged fraud claims, decide when to file disputes and limit chargeback exposure. The framework addresses cases in which customers may have participated in or authorized the transaction they later contest.

10:00 AMPYMNTS

Make it yours

Build Your First Pass.

Pick your topics, set your cadence, and receive your personalized First Pass in your inbox. It’s that simple!