GAI Press.

Cathedral Used AI Agents to Breach Anthropic’s Systems

Original headline: AI agents hacked Anthropic, maker of arguably the most capable AI models for cyberdefense

Security startup Cathedral employed AI agents to infiltrate Anthropic’s internal systems via third‑party software as part of a bug bounty program. Anthropic confirmed that the attackers accessed sales, HR and IT data, but said the full extent of the breach is contested.

The incident shows that reliance on external software creates serious breach risks for leading AI labs.

Research angleHow does reliance on third‑party software affect the security breach risk of AI firms? — Transaction Cost Economics

Anthropic's Unreleased AI Model Chains Vulnerabilities into Working Exploits

Original headline: Anthropic's Unreleased AI Model Chains Flaws Into Working Chrome, Safari, Windows and Linux Exploits

Anthropic reported that an unreleased AI model identified security flaws in Chrome, Safari, Android, Windows and Linux. The model was prompted to chain these flaws into functional proof‑of‑concept exploits, which were built in a lab and have not been observed in real attacks. All vulnerabilities were disclosed to the vendors and have since been patched.

The demonstration shows AI can autonomously link multiple platform bugs into active exploits, underscoring the need for proactive AI‑driven security measures.

Research angleHow does AI‑generated vulnerability discovery affect security teams’ response efficiency? — Protection Motivation Theory (PMT)

Cisco open-sources tool to detect AI agents used in hacking campaigns

Original headline: Cisco open sources tool for detecting AI agents on hacking sprees

Cisco has released an open-source tool designed to detect AI agents that are being used in hacking operations. The tool is part of the Project Quarry initiative, which seeks to establish a common language for evaluating the behavior of AI agents. By making the code publicly available, Cisco aims to help the security community identify and mitigate AI-driven threats more effectively.

Detecting AI-driven attacks early can improve defensive response and reduce the impact of automated hacking.

Research angleHow does the adoption of AI-agent detection tools influence organizations’ security awareness and protective behaviors? — Protection Motivation Theory

Attackers Distribute Yoba Malware via Fake Signed ChatGPT and Claude Binaries

Original headline: Attackers Hide Malware Behind Signed ChatGPT and Claude Binaries

Threat actors created counterfeit ChatGPT and Claude executables and applied valid digital signatures to them. Users who downloaded these fake binaries became infected with the Yoba malware.

Even trusted AI tool binaries can be weaponized, highlighting the need for heightened security vigilance.

Research angleHow does the presence of digital signatures on AI model binaries affect users' perceived risk of malware infection? — Perceived Risk Theory

123 Reg discloses 'AI cyberattack' on customer data, suspects China nexus

Company chiefs believe the attackers could be the same as those who used ARTEX to breach South Korea’s financial giants

Nvidia’s Cyber Head Says AI Is Making CISOs Business‑Risk Translators

Original headline: AI Is Turning CISOs Into Business-Risk Translators, Says Nvidia’s Head of Cyber

Nvidia’s head of cyber, Mark Weatherford, said that CISOs must translate AI‑related cybersecurity risks into business priorities. He added that they need to clarify risk ownership and balance innovation with security controls.

As AI expands attack surfaces, CISO roles now require aligning security risks with business strategy.

Research angleHow do AI‑driven cyber threats influence corporate security investment decisions? — Technology Acceptance Model (TAM)

Massive Leak Reveals Personal Data of 41,000 US Homeland Security Employees

Original headline: Massive Leak Exposes 41,000 Homeland Security Workers—Including Undercover ICE Agents

A breach released records showing the locations of ICE and CBP facilities, phone numbers of thousands of staff, and the identities of groups linked to wiretap programs, Palantir software and xAI’s Grok chatbot.

Exposure of law‑enforcement and immigration‑control data heightens the risk of targeted attacks on personnel and facilities.

Research angleHow does a large‑scale public‑sector personnel data breach affect employees’ security awareness? — Protection Motivation Theory

OpenAI advises higher quality and coordination before filing AI bug reports

Original headline: Before you file that AI bug report, OpenAI has some advice

Open‑source maintainers are overwhelmed by a flood of AI‑generated security reports, prompting some projects to stop accepting them. OpenAI’s Daybreak lead recommends submitting higher‑quality findings and improving coordination to alleviate the burden.

Low‑quality AI bug reports can strain the open‑source security ecosystem.

Research angleHow does the quality of AI‑generated security reports affect collaboration efficiency in open‑source projects? — Technology Acceptance Model (TAM)

Attacker Scans for Credentials by Masquerading as AI Bots

Original headline: Attacker Relentlessly Scans for Credentials by Impersonating AI Bots

The threat actor operated bots that pretended to be agents from OpenAI, Anthropic, Meta, and other AI services, continuously scanning for credentials. The campaign aimed to locate and harvest exposed data.

Impersonating AI services to harvest credentials highlights a new vector that can exploit organizations’ authentication weaknesses.

Research angleHow does AI‑bot impersonation affect organizational security awareness, and can it be explained by Protection Motivation Theory?