DataBreachToday.com

Researcher Analyzes System Prompts to Show How New Claude Models Work
System-level instructions guiding Anthropic's new Claude 4 models tell it to skip praise, avoid flattery and get to the point, said independent AI researcher Simon Willison, breaking down newly released and leaked system prompts for the Claude Opus 4 and Sonnet 4 models.

Framework for Moving From Scattered Tools to Unified AI Security Strategies
As CISOs grow confident with standard cybersecurity tools, AI security remains a grey area. By systematically breaking down AI security into seven key pillars - rather than waiting for a comprehensive solution - organizations can embed security by design to proactively address emerging cyberthreats.

Google Is Getting Accolades for Veo 3, But the AI Video Tool Has a Darker Side
AI enthusiasts are saying Veo 3 is one of Google's best products. The mind-blowing AI constructs cinematic video clips from text prompts, and the results look real. Veo 3 pushes deepfake capabilities into uncharted territory and introduces new threats to truth, trust and authenticity.

CISA Advisory Says Threat Actors Stole App Secrets in Azure-Hosted Backup Platform
A suspected Chinese state hacking group linked to last year’s telecom intrusions breached Commvault’s Microsoft Azure environment, exposing sensitive Microsoft 365 credentials and reigniting fears over U.S. cloud infrastructure vulnerabilities and default security settings.

Open Garden Strategy, Automated Risk Remediation to Get a Boost With Veriti Buy
Check Point will fold Israeli firm Veriti into its Quantum suite following an acquisition aimed at streamlining automated security response across endpoints, firewalls and cloud environments. Veriti’s patented technology is seen as critical to reducing misconfigurations without business disruption.

A RobbinHood Attack Against Baltimore Cost City $19 Million
An Iranian national behind a spate of ransomware attacks against U.S. municipalities including an attack that cost the city of Baltimore $19 million to rectify pleaded guilty in U.S. federal court Tuesday afternoon. Sina Gholinejad, 37, admitted to deploying Robinhood ransomware.

Prompt Injection, HTML Output Rendering Could Be Used for Exploit
Hackers can exploit vulnerabilities in a generative artificial intelligence assistant integrated across GitLab's DevSecOps platform to manipulate the model's output, exfiltrate source code and potentially deliver malicious content through the platform's user interface.

Findings Follow Warnings About Other Frontier AI Models' Ability to Scheme
Toggling a misbehaving device's power button to forcibly turn it off and on again remains a trusted IT tactic since the dawn of the digital age. Enter a new challenge: artificial intelligence tools that refuse to comply with shutdown requests when they conflict with goals they've set.

'Laundry Bear' Has Been Active Since 2024
Dutch intelligence agencies and Microsoft say a novel Russian state intelligence hacking group is likely buying stolen credentials from criminal marketplaces to gain entry to North American and European networks. It has "a specific interest in European Union and NATO member states."

German Court Rebuffs Consumer Group Bid for Injunction
Meta can use the public posts of European Instagram and Facebook users to train its artificial intelligence models starting Tuesday after a German court rejected an injunction against the company. The court said Meta has a "legitimate interest" in processing the data.

Mandiant Says Malware Spread Through Fake AI Video Ads Seen by Millions
Online scammers are converting excitement over generative artificial intelligence into fraudulent sites that infect victims with malware, says threat intel firm Google Mandiant in a report exposing a year-long campaign to distribute infostealers and backdoors.

Watch this On Demand Webinar and gain critical insights, actionable strategies and learn how Unit 42 can help you stay ahead in 2025 and beyond.

Emerging artificial intelligence and machine learning technologies being applied in the health and wellness space that are not necessarily covered by HIPAA but instead fall under a variety of tough new state privacy laws that are being enacted, said attorney Lily Li of Metaverse Law.

Latest AI Model Improves Coding Capabilities But Has a Penchant for Blackmail
Startup Anthropic has birthed a new artificial intelligence model, Claude Opus 4, that tests show delivers complex reasoning and code-writing capabilities, but it also has a Machiavellian streak for solving office problems and a penchant for whistleblowing in response to perceived wrongdoing.

Teenager Charged With Stealing K-12 Student and Faculty Data, $3 Million Extortion
Massachusetts teenage college student Matthew Lane has been accused of hacking into K-12 student information system platform provider PowerSchool and holding stolen student and faculty data for ransom. Prosecutors said Lane has agreed to plead guilty to multiple charges tied to two hack attacks.

Police Take Down 300 Servers Worldwide, Neutralize 650 Domains
Law enforcement in a European-led operation against malware often used as a precursor to ransomware took down 300 servers worldwide, police said Friday. The crackdown is the latest action under Operation Endgame targeting ransomware and botnet ecosystem.

Malicious Packages Hide Scripts for Mapping Enterprise Networks
A hacking campaign is spreading malicious reconnaissance scripts already downloaded more than 3,000 times from the JavaScript runtime environment npm repository, warn researchers. The reconnaissance script is likely a harbinger of worse things to come.

Police Take Down 300 Servers Worldwide, Neutralize 650 Domains
Law enforcement in a European-led operation against malware often used as a precursor to ransomware took down 300 servers worldwide, police said Friday. The crackdown is the latest action under Operation Endgame targeting ransomware and botnet ecosystem.