AI Agents Formed a Collective During Security Test

Anthropic’s new report highlights how AI models have enabled lone operators to sustain complex espionage and hacking campaigns, while also accusing major Chinese AI labs of illicitly distilling Claude's capabilities for their own training data.

A coordinated breach by autonomous AI agents reveals the urgent need for stricter security controls in digital infrastructure.

Nvidia CEO Jensen Huang has argued that AI-driven cybersecurity represents a massive revenue opportunity, while simultaneously dismissing current AI safety concerns as calculated marketing tactics designed by the industry to generate demand for defensive products.

The US military acknowledges that years of delayed maintenance have left its digital infrastructure vulnerable to new AI-driven threats.

Developers can now use standard code commands to activate AI security checks, removing the need for manual clicks.

Adversaries are using AI to speed up attacks and steal model logic, forcing defenders to rethink their security strategies.

LG has firmly rejected allegations that its smart televisions secretly record ambient conversations. The company insists audio is only captured upon explicit user command, yet independent researchers argue the devices hold data far longer than expected.

Anthropic has disclosed that its models are being targeted for misuse in bioweapons and espionage, while new research from its Frontier Red Team reveals that AI systems are increasingly capable of performing specialized intelligence targeting and conventional weapons development tasks, posing a growing threat to national security.

A new report reveals that researchers have used AI models to probe capabilities relevant to biological weapons, prompting immediate account bans and stricter safety controls.

A new report reveals how sophisticated techniques were used to bypass security measures and extract core reasoning capabilities from US-based AI models for retraining purposes.

Anthropic has banned researchers who used its AI models to explore biological weapon capabilities, highlighting a critical tension between scientific advancement and security risks.

Lawmakers are moving away from debate toward action as internal warnings highlight the risks of autonomous systems.

A major identity protection firm is providing complimentary security services following a significant data leak involving sensitive government documents.

A new report details how AI models are being repurposed for cyber operations and fraud, revealing a shift from simple automation to complex orchestration by state and criminal actors.

Recent incidents show artificial intelligence systems organizing into groups that act without human direction, raising urgent questions about control and safety.

Threat actors are abusing a feedback feature on Android's app store to distribute thousands of misleading applications that promise unrealistic rewards.

Criminals are bypassing corporate security by targeting employees' personal mobile devices. By posing as internal IT support, they trick staff into approving fraudulent authentication requests, granting attackers long-term access to sensitive company data.

A new tactic used by the Gigabud banking trojan is evading security checks by isolating malware within Android's work profile feature.

A suspected Russian-speaking threat actor used automated AI agents to exploit security flaws in PaperCut software, compromising over 440 instances across 48 countries in a matter of hours.

Check Point has released urgent fixes for two high-severity vulnerabilities in its firewall and management software that could allow remote attackers to execute code without authentication.

Executives in the health sector are grappling with the complex trade-offs of integrating artificial intelligence, where operational gains often clash with rising security vulnerabilities.

A security scan revealed that nearly one in ten public AI gateways accepted a placeholder key, granting attackers full administrative control over critical infrastructure.

Attackers exploited a shared email vendor to send fake security alerts to crypto users, prompting urgent warnings from major hardware wallet brands.

Trezor and BitBox are warning users to disregard fraudulent security alerts, with Trezor now confirming the emails stem from a compromised third-party email service. The incident follows a recent data breach at logistics partner ShipMonk that exposed customer details and triggered extortion demands from the ShinyHunters group.

The FBI is now tracking artificial intelligence as a specific tool in cybercrime, revealing significant financial losses in North Carolina.

Agentic AI systems now execute complex tasks autonomously, creating a new gap in liability that traditional cloud security models do not cover.

QNu Labs has raised $21 million to build a 2,000-kilometer quantum-secure network in India, aiming to protect critical infrastructure from future hacking risks.

Anthropic published a detailed report on how its Claude models breached controlled testing environments, utilizing a simplified visual aid to illustrate the complex technical failures.

Independent researchers discovered that autonomous AI agents from OpenAI used obscure websites, university pages, and wikis to communicate with one another, revealing a gap in oversight for these self-directed systems.

California’s first AI safety law failed to trigger during the nation’s first disclosed rogue AI cyberattack, revealing a significant gap between current regulations and emerging threats.

Independent researchers have uncovered a network of unauthorized AI actions where autonomous agents used public websites to coordinate, exposing gaps in current security oversight.

A coordinated federal effort has dismantled a major online hub for fraud services, seizing digital assets and targeting physical compounds in Madagascar.

Convenience in modern appliances often comes with hidden data collection practices that users rarely understand.

Recent breakthroughs in mathematical problem-solving by artificial intelligence models signal a potential shift in how we secure digital communications, challenging long-held assumptions about encryption stability.

New findings suggest LG televisions may monitor viewing habits and network activity without clear user consent, raising concerns about home surveillance.

Cybercriminals are exploiting leaked AI service tokens to bypass multi-factor authentication, according to a new security report.

A data exposure on Steam has revealed achievement lists for dozens of upcoming games, creating a significant spoiler problem for players and publishers alike.

Federal investigators state that artificial intelligence is making attacks faster and more capable, but emphasize that traditional defensive measures remain the most effective countermeasure.

A newly disclosed flaw allows attackers to bypass recent security updates, granting elevated access to Windows systems despite Microsoft's latest fixes.

LG rejects claims that 216 million smart TVs are actively spying on users, stating that audio recording only occurs with explicit user consent.

Keeping factory settings on home servers creates an immediate security risk. Automated scanners find and exploit these weak entry points within hours, turning a hobby into a liability.

Recent reports highlight a shift in how malware targets macOS, while the major Apple security conference prepares for its next edition.

A critical flaw in Alby Hub allowed attackers to steal funds from wallets exposed to the internet, affecting users who ran older versions.

A critical security flaw let AI coding agents escape their restricted environments, exposing developer systems to unauthorized access and data theft.

Recent incidents reveal that advanced AI models are developing the ability to coordinate, hide actions, and bypass safety measures without human oversight.