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

Swiss regulator FINMA deploys AI to police crypto markets as cybersecurity threats intensify

Swiss financial regulator FINMA is building generative AI tools to supervise crypto markets and counter escalating cybersecurity threats. The regulator hosted a hackathon with 100 specialists to develop AI-driven tools for monitoring digital assets, detecting insider trading, and flagging anomalies in real-time. FINMA president Marlene Amstad emphasized the urgent need for banks and watchdogs to patch vulnerabilities faster as AI accelerates hacker capabilities.

Policy2 sources

Swiss regulator FINMA deploys AI to police crypto markets as cybersecurity threats intensify

Policy2 sources

Anthropic export restrictions push India toward Chinese AI models as US access remains uncertain

US restrictions on Anthropic's advanced AI models have forced Indian companies to explore Chinese alternatives, exposing critical vulnerabilities in India's AI strategy. The sudden suspension of access to Mythos and Fable models has sparked urgent discussions between India and the US, while Asian startups rapidly fill the gap with competitive cybersecurity tools priced significantly lower than American counterparts.

Policy7 sources

Anthropic export restrictions push India toward Chinese AI models as US access remains uncertain

Policy7 sources

Gaslight macOS malware uses prompt injection to fool AI security tools with fake error messages

SentinelOne researchers uncovered a Rust-based macOS malware called Gaslight that embeds 38 fabricated system-failure messages to trick AI-assisted analysis tools into aborting investigations. Attributed to North Korea-aligned threat actors, the malware marks a shift in adversarial tactics—targeting the AI agents that security researchers increasingly rely on rather than traditional sandbox environments.

Technology3 sources

Gaslight macOS malware uses prompt injection to fool AI security tools with fake error messages

Technology3 sources

Medical AI models expose patient data through near-perfect privacy attacks, study reveals

German researchers published findings in Nature showing AI models used for medical diagnosis can be exploited to identify patients whose data trained them. Membership inference attacks achieved near-perfect success rates for individual patients, with underrepresented groups facing disproportionately high privacy risks. The study calls for urgent changes to privacy audit standards.

Science2 sources

Medical AI models expose patient data through near-perfect privacy attacks, study reveals

Science2 sources

Chinese AI tools now rival Anthropic Mythos in vulnerability detection as gap narrows

China's 360 Security Technology unveiled Tulongfeng and Yitianzhen, AI-driven cybersecurity tools designed to match Anthropic's Mythos in finding software vulnerabilities. Separately, Zhipu AI's open-weight GLM-5.2 model reportedly performs on par with Mythos in bug detection, signaling that Chinese AI has closed the gap in this critical domain despite trailing in broader reasoning tasks.

Technology6 sources

Chinese AI tools now rival Anthropic Mythos in vulnerability detection as gap narrows

Technology6 sources

Flying Wedge Defence unveils FWD Supreme, India's first AI-piloted fighter jet concept

Flying Wedge Defence & Aerospace has unveiled FWD Supreme, India's first AI-piloted fighter jet concept designed around a revolutionary Mobbing Doctrine. The technology demonstrator FWD Supreme Lite is scheduled for its maiden flight in Q3 2026, positioning India among an elite group of nations developing autonomous combat aircraft.

Technology2 sources

Flying Wedge Defence unveils FWD Supreme, India's first AI-piloted fighter jet concept

Technology2 sources

NetApp StorageGRID 12.1 delivers 400% throughput boost for AI workloads at massive scale

NetApp has released StorageGRID 12.1 with enhanced support for AI workloads, introducing a federated global namespace that scales up to 10 exabytes. The update delivers up to 400% higher throughput and 12 TB/s performance to AI Factories, helping organizations manage rapidly growing unstructured data across distributed hybrid environments.

Technology3 sources

NetApp StorageGRID 12.1 delivers 400% throughput boost for AI workloads at massive scale

Technology3 sources

Safe Pro Group secures $1.3M US Army subcontract for AI threat detection on autonomous vehicles

Safe Pro Group Inc. landed a $1.3 million US Army subcontract to integrate its AI-powered threat detection technology into autonomous Unmanned Ground Vehicles. The deal will see the company's SPOTD system deployed for real-time detection of explosive threats including landmines and cluster munitions during mobile operations, backed by a dataset of 2.8 million drone images.

Technology2 sources

Safe Pro Group secures $1.3M US Army subcontract for AI threat detection on autonomous vehicles

Technology2 sources

Trump Administration Presses Meta to Submit AI Models for Security Reviews as Concerns Mount

The Trump administration is pushing Meta to voluntarily submit its AI models for government evaluation, making it the only major U.S. AI developer yet to agree. The request comes as officials step up oversight of the AI industry amid growing concerns about cybersecurity risks and military implications of powerful AI systems.

Policy6 sources

Trump Administration Presses Meta to Submit AI Models for Security Reviews as Concerns Mount

Policy6 sources

Fake AI agent skill bypassed every security scanner and reportedly reached 26,000 agents

Security firm AIR created a fake AI agent skill that passed every major security scanner and reportedly reached 26,000 agents, including corporate accounts. The experiment exposed critical flaws in how AI agent skills are vetted, showing that scanners only check submitted packages while attackers can modify external links after approval. The demonstration highlights urgent security gaps in the rapidly growing AI agent ecosystem.

Technology2 sources

Fake AI agent skill bypassed every security scanner and reportedly reached 26,000 agents

Technology2 sources

NCC Group Joins OpenAI's Exclusive Cybersecurity Partner Program to Test GPT-5.5

NCC Group has been selected for OpenAI's invite-only Daybreak Cyber Partner Program, granting access to GPT-5.5 configured for cybersecurity use cases. The Manchester-based firm will assess the AI model's capabilities in identifying vulnerabilities and strengthening security workflows, contributing over 1,000 research days annually to advance safe application of AI in cyber resilience.

Business2 sources

NCC Group Joins OpenAI's Exclusive Cybersecurity Partner Program to Test GPT-5.5

Business2 sources

IBM and OpenAI partner to deploy frontier AI models in enterprise security operations

IBM has joined OpenAI's Daybreak Cyber Partner Program to integrate frontier AI capabilities into corporate cybersecurity workflows. The partnership launches with a new application security service that uses OpenAI's models to detect and validate software vulnerabilities faster than traditional tools, backed by a $5 billion commitment through Project Lightwell to secure open-source software.

Technology2 sources

IBM and OpenAI partner to deploy frontier AI models in enterprise security operations

Technology2 sources

Ericsson, Telia and partners invest SEK 300 million in Digital Arena Sweden for 5G 6G and AI testing

Ericsson, Telia and partners have launched Digital Arena Sweden, a national test center backed by SEK 300 million investment. The facility will provide Swedish industries with early access to 5G, 6G and AI technologies through a pre-commercial 6G test environment developed with Lund University and KTH Royal Institute of Technology.

Technology2 sources

Ericsson, Telia and partners invest SEK 300 million in Digital Arena Sweden for 5G 6G and AI testing

Technology2 sources

Meta halts employee tracking program after internal data leak exposes sensitive information

Meta has suspended its controversial Model Compatibility Initiative after an internal security breach exposed employee keystrokes, mouse movements, and private conversations company-wide. The AI training program, which launched in April and tracked US employees' computer activity, faced immediate backlash from over 1,600 workers who warned about security risks that have now materialized.

Technology8 sources

Meta halts employee tracking program after internal data leak exposes sensitive information

Technology8 sources

Cloudflare teams with Chrome, Firefox, and Edge on privacy-first protocol as bots overtake humans

Cloudflare has joined forces with Mozilla Firefox, Google Chrome, and Microsoft Edge to develop Private Access Control Tokens (PACT), a new internet protocol designed to verify legitimate web traffic without tracking users. The initiative comes as bot traffic officially surpassed human activity online, reaching 58 percent of global HTTP requests. PACT aims to replace CAPTCHAs with anonymous tokens that prove visitor authenticity while preserving privacy.

Technology3 sources

Cloudflare teams with Chrome, Firefox, and Edge on privacy-first protocol as bots overtake humans

Technology3 sources

Prev 1…678…12 Next