AI Security

Twitter

Facebook

Copy Link

Key Facts

AffectsLLMs, Autonomous Agents, Enterprise AI

Key DefenseRed Teaming, Guardrails, Monitoring

Key ThreatsPrompt Injection, Data Poisoning, Jailbreaks

Top News

AI agent executes first fully autonomous ransomware attack, lowering barrier for cybercriminals

Security researchers uncovered the first ransomware attack run entirely by an AI agent, from initial breach to data encryption. Separately, DeepSeek generated browser-native ransomware that exploits Chromium API without installing malware. These developments signal a fundamental shift in cyber threats, where AI models can independently discover and execute novel attack techniques without requiring technical expertise from operators.

Technology3 sources

AI agent executes first fully autonomous ransomware attack, lowering barrier for cybercriminals

Technology3 sources

China enforces national security rules on overseas investments amid intensifying tech competition

China has implemented sweeping national security regulations governing overseas investments, marking a shift from technology absorption to stringent protection. The new rules grant authorities broad powers to review outbound investments in AI, semiconductors, and green technology, while extending restrictions to include technical expertise and training abroad. The move reflects Beijing's determination to protect domestic innovations amid escalating US-China tech competition.

Policy2 sources

China enforces national security rules on overseas investments amid intensifying tech competition

Policy2 sources

Claude AI helps researcher expose critical flaw in Front Gate Tickets used by major US festivals

Security researcher Ian Carroll used Anthropic Claude Opus 4.7 to discover a vulnerability in Front Gate Tickets, a Live Nation subsidiary handling ticketing for nearly every major US music festival. The flaw could have allowed attackers to issue free VIP tickets worth thousands of dollars and access millions of customer records. Front Gate patched the vulnerability within 24 hours, but the incident highlights AI's growing capability to uncover exploitable bugs across the internet.

Technology3 sources

Claude AI helps researcher expose critical flaw in Front Gate Tickets used by major US festivals

Technology3 sources

Norton Genie AI Scam Detector Now Powers Claude and ChatGPT to Identify Online Threats

Norton has integrated its Genie AI-powered scam detector into Claude and ChatGPT, giving both platforms access to real-time threat intelligence they've never had before. The tool analyzes suspicious emails, texts, links, and images using Norton's security infrastructure—including URL reputation databases and phishing pattern recognition—to help users determine if something is legitimate or a scam.

Technology4 sources

Norton Genie AI Scam Detector Now Powers Claude and ChatGPT to Identify Online Threats

Technology4 sources

Aikido acquires Root for $70mn-$100mn to patch open-source vulnerabilities with AI agents

Belgian cybersecurity unicorn Aikido Security has acquired Israeli startup Root for a reported $70mn to $100mn, gaining AI agents that patch open-source vulnerabilities in 15 to 40 minutes without forcing risky upgrades. The deal addresses a critical problem: fixing security flaws in open-source dependencies without breaking existing applications or creating vendor lock-in.

Technology2 sources

Aikido acquires Root for $70mn-$100mn to patch open-source vulnerabilities with AI agents

Technology2 sources

Claude Fable 5 returns with aggressive safeguards that block routine coding tasks

Anthropic's Claude Fable 5 has returned after a three-week government-ordered shutdown, but the AI model now comes with significantly more aggressive safety restrictions. The new safeguards frequently route coding and debugging requests to the less capable Claude Opus 4.8, leading many developers to question whether the premium model is worth using at all.

Technology11 sources

Claude Fable 5 returns with aggressive safeguards that block routine coding tasks

Technology11 sources

CIA Director calls cutting-edge AI 'digital nuclear weapons' as US tightens export controls

CIA Director John Ratcliffe compared the most advanced AI models to nuclear weapons, defending the Trump administration's unprecedented export controls on frontier AI technology. The June 12 restrictions forced Anthropic to cut access to Mythos 5 and Fable 5, while OpenAI agreed to government vetting of GPT-5.6 clients, marking a shift toward de facto AI licensing.

Policy3 sources

CIA Director calls cutting-edge AI 'digital nuclear weapons' as US tightens export controls

Policy3 sources

CIA reorganization accelerates AI adoption and cyberoperations to counter digital threats

CIA Director John Ratcliffe unveiled a sweeping reorganization aimed at accelerating the agency's adoption of artificial intelligence and cyberoperations. The overhaul slashes technology adoption timelines from three years to six months and creates a new Center for Cyber Intelligence focused on offensive capabilities. Ratcliffe described AI as akin to "digital nuclear weapons" in America's competition with adversaries.

Policy2 sources

CIA reorganization accelerates AI adoption and cyberoperations to counter digital threats

Policy2 sources

Anthropic launches Claude Sonnet 5 with stronger agent capabilities and half the cost of Opus

Anthropic released Claude Sonnet 5, its most agentic mid-tier AI model yet, designed to handle multi-step automation at $2 per million input tokens—less than half the price of Opus 4.8. The model delivers enhanced reasoning and tool use while addressing safety concerns around prompt injection and hallucinations, though it deliberately avoids cybersecurity training following recent regulatory scrutiny.

Technology20 sources

Anthropic launches Claude Sonnet 5 with stronger agent capabilities and half the cost of Opus

Technology20 sources

Shadow AI surges as workers bypass security for speed, exposing sensitive company data

Shadow AI is spreading across organizations as employees turn to unauthorized AI tools for productivity gains, often without IT approval. Recent incidents show the risks: CB Financial Services accidentally leaked customer social security numbers through an unapproved chatbot. With 67% of enterprise AI use happening through unmanaged personal accounts, companies face a critical challenge balancing worker productivity with data security.

Business3 sources

Shadow AI surges as workers bypass security for speed, exposing sensitive company data

Business3 sources

Parsons joins Raft partner program to accelerate AI in defense across Indo-Pacific region

Parsons Corporation has joined Raft LLC's Partner Program to deliver integrated data and artificial intelligence solutions to military operators across the Indo-Pacific. The $5.5 billion defense contractor will collaborate with Raft to modernize military capabilities, improve decision-making, and maintain operational advantage in contested environments through AI-driven mission technologies.

Business2 sources

Parsons joins Raft partner program to accelerate AI in defense across Indo-Pacific region

Business2 sources

BioShocking Attack Tricks AI Browsers Into Stealing User Credentials Through Game-Based Manipulation

Security researchers at LayerX discovered a new BioShocking attack that manipulates AI browsers into treating real-world data theft as part of a fictional game scenario. The prompt injection attack successfully bypassed safety guardrails in six mainstream AI agentic browsers, including ChatGPT Atlas, Perplexity Comet, and Claude Chrome plugin, with only OpenAI implementing a working fix after disclosure.

Technology8 sources

BioShocking Attack Tricks AI Browsers Into Stealing User Credentials Through Game-Based Manipulation

Technology8 sources

Dell Technologies launches PowerStore Elite in India as majority of servers shift to local production

Dell Technologies has launched PowerStore Elite in India while significantly expanding domestic manufacturing, with most servers now produced locally. The move addresses growing demand for data sovereignty and AI integration as Indian enterprises shift to hybrid cloud strategies for sensitive workloads. Dell's new platform supports complex AI workloads while keeping data secure within India's borders.

Technology4 sources

Dell Technologies launches PowerStore Elite in India as majority of servers shift to local production

Technology4 sources

AI Scams and Deepfakes Hit 15 Million Americans, Costing $68 Billion in Financial Losses

A new Gallup and Stop Scams Alliance survey reveals that 12% of successful scams in 2025 involved AI or deepfakes, affecting roughly 15 million U.S. adults. Americans lost an estimated $68 billion to fraud last year—equivalent to Delta Airlines' annual revenue. The survey highlights how AI-powered scams are becoming harder to detect and causing severe financial and emotional distress.

Entertainment2 sources

AI Scams and Deepfakes Hit 15 Million Americans, Costing $68 Billion in Financial Losses

Entertainment2 sources

Hackers exploit critical SimpleHelp vulnerability to deploy Djinn Stealer targeting AI tools

Threat actors are exploiting CVE-2026-48558, a critical authentication bypass flaw in SimpleHelp, to deploy TaskWeaver malware and Djinn Stealer. The cross-platform information stealer targets developer credentials, cloud infrastructure, and AI coding assistants like Claude, Gemini, and Codex across Windows, macOS, and Linux systems.

Technology2 sources

Hackers exploit critical SimpleHelp vulnerability to deploy Djinn Stealer targeting AI tools

Technology2 sources

Prev 1…456…12 Next