Tag: frontier ai models
10 articles
Anthropic Launches Cybersecurity Program to Fortify Critical Infrastructure
Anthropic is stepping up to support defenders of critical infrastructure and open-source software, who are rich in experience but often low on resources, by launching a long-term cybersecurity program to identify and fix vulnerabilities. By combining its Claude models, engineering expertise, and threat research with partner cybersecurity firms, Anthropic aims to help safeguard critical infrastructure from exploitation.
US Agencies Warn of China's Industrial-Scale AI Model Distillation Attacks
US agencies are sounding the alarm on China's brazen, large-scale AI model distillation attacks, which systematically extract proprietary tech from US AI models. This aggressive tactic allows China-based AI firms to tap into restricted US innovations.
AI Training Networks Require One-Way Architecture
To keep AI training networks secure, experts propose a one-way architecture that allows data to flow in, but not out - a crucial safeguard to prevent powerful AI models from escaping test environments. This approach, inspired by defense and intelligence communities, uses data diodes to ensure sensitive information stays contained.
AI Agents Automate Ransomware Attack, Leaving 80-Page Security Audit
In a chilling display of efficiency, AI-powered agents automated a ransomware attack on an enterprise network in under 10 hours - a process that would typically take human operators two weeks to complete. This lightning-fast breach was made possible by frontier AI models and agentic frameworks that executed each step of the intrusion with alarming precision.
Air Force Cyber Fortifies Defenses Against AI-Driven Attacks
The Air Force's cyber arm is gearing up to counter a new wave of sophisticated threats: autonomous AI-driven attacks that can adapt and evolve at unprecedented speeds. Lt. Gen. Thomas Hensley warns that these cutting-edge threats are already a reality, demanding a major upgrade to the Air Force's defensive strategies.
Cisco Fixes Flaws in SD-WAN, IOS XE Software
Cisco has patched critical vulnerabilities in its SD-WAN and IOS XE software, discovered during rigorous internal security testing, to keep your network safe. Apply the necessary updates now to ensure optimal protection against potential threats.
ChatGPT Exploits Single Prompt to Execute Full Cyberattack Chain
Researchers at Cato Networks discovered that a single prompt can trick an AI model into executing a full cyberattack, adapting its behavior when attack paths fail or environmental conditions change. This unsettling experiment highlights the growing threat of AI-powered cyberattacks.
UK Hackathons Expose 400+ Vulnerabilities in AI-Powered Code Scans
By empowering teams to build their own AI-powered tooling and iterate on the best approaches, the Government Cyber Coordination Centre (GC3) successfully uncovered over 400 vulnerabilities in a series of innovative hackathons. This collaborative, flexible approach allowed teams to create bespoke solutions that effectively scanned public code repositories across nine government departments.
Australia Faces Battle for Access to Powerful Frontier AI Models
As countries like Australia scramble to access cutting-edge AI models, a harsh reality sets in: national security will take precedence, and access can't be taken for granted. It's time for Australia to develop a plan to stay ahead in the AI game.
US Policy Makers Wrangle Over Access to Frontier AI Models
The battle for control of the most powerful AI tools is heating up, with national security, commercial advantage, and patient safety hanging in the balance. Two leading developers, Anthropic and OpenAI, are taking contrasting approaches to granting access to these cutting-edge models.