Anthropic's Claude Mythos AI system has made significant strides in identifying software vulnerabilities, uncovering over 10,000 severe vulnerabilities across various digital infrastructure systems as part of Project Glasswing. This includes 6,202 high- or critical-severity vulnerabilities affecting over 1,000 open-source software projects.
Researchers testing Anthropic's Mythos and OpenAI's GPT-5.5 have found that these AI models possess advanced hacking capabilities that could potentially be used for widespread destruction. Experts warn that the impact could be larger than imagined, prompting entities like the National Cyber Security Centre (NCSC) in New Zealand to engage with tech companies and learn from US firms.
Google has taken a different approach, showcasing practical AI applications at its recent I/O event. However, concerns about trusting AI tools persist. In response to AI-related risks, Pope Leo has called for stronger oversight and disarmament of AI-powered weapons systems, emphasizing the need for humans to maintain control over AI systems.
Security researchers have also demonstrated a new type of attack, called AudioHijack, which uses hidden audio signals to manipulate voice assistants into carrying out unauthorized actions. Meanwhile, companies like ServiceNow are leveraging AI for workflow efficiency, and Google has introduced an open-source Agent Executor to support AI agents in production.
The increasing use of AI in various sectors has raised concerns about cyber security. The European Central Bank (ECB) has urged banks to strengthen their cyber security systems after new AI models exposed weaknesses. Additionally, a growing share of hourly workers in the US are encountering AI on the job, highlighting the need for better preparation and understanding of AI's impact.
Key Takeaways
• Anthropic's Claude Mythos AI system identified over 10,000 severe software vulnerabilities as part of Project Glasswing. • Researchers found advanced hacking capabilities in Anthropic's Mythos and OpenAI's GPT-5.5. • Google showcased practical AI applications at its I/O event. • Pope Leo called for stronger oversight and disarmament of AI-powered weapons systems. • Security researchers demonstrated a new type of attack, AudioHijack, targeting voice assistants. • ServiceNow is using AI for workflow efficiency. • Google introduced an open-source Agent Executor to support AI agents in production. • 65% of decision-makers use shadow AI in the workplace. • 37% of hourly workers in the US encounter AI on the job. • The ECB urged banks to strengthen cyber security systems due to AI-related risks.AI Models Jolting Washington with Advanced Hacking Capabilities
Researchers who have tested Anthropic's Mythos and OpenAI's GPT-5.5 say these AI models have advanced hacking capabilities that could change the digital security landscape forever. These tools are advancing much faster than anticipated and could potentially be used to create widespread destruction. Experts warn that the fallout could be larger than imagined.
Anthropic's Claude Mythos AI Uncovers Over 10,000 Vulnerabilities
Anthropic's Claude Mythos AI system has helped identify over 10,000 severe software vulnerabilities across various digital infrastructure systems. The initiative, called Project Glasswing, has already uncovered 6,202 high- or critical-severity vulnerabilities affecting over 1,000 open-source software projects. This development signals a new era in defensive cyber operations.
New Zealand at the Forefront of AI Superhacking
New Zealand's cyber watchdog is learning from US companies testing superhacking AI models that have sent US banks rushing to plug holes. The National Cyber Security Centre (NCSC) in New Zealand has warned of a significant increase in vulnerabilities and incidents. The NCSC is engaging with tech companies, including developers of frontier models and companies that have had access to preview models like Mythos.
Pope Leo Calls for Disarming of Artificial Intelligence
Pope Leo has warned that some AI-powered weapons systems are becoming practically beyond human control. He called for stronger oversight in his long-awaited manifesto on the rapidly developing technology. The Pope said the Catholic Church wants to work with AI developers to discuss the proper use of the technology.
Pope Warns of AI Weapons Beyond Human Control
Pope Leo warned that some autonomous weapons systems are practically beyond human reach. He presented his encyclical on the risks of AI systems and emphasized the need for humans to maintain key controls of AI systems.
Google Showcases Practical AI Applications
Google's recent I/O event highlighted practical AI applications, but the author still has mixed feelings about the future of AI. While Google shows practical and useful ideas for AI, the author worries about trusting AI tools. The author believes AI should be built with humans in mind.
AI Voice Bots Vulnerable to Hidden Audio Attacks
Security researchers have demonstrated a new type of attack that uses hidden audio signals to manipulate voice assistants into carrying out unauthorized actions. This proof-of-concept technique, called AudioHijack, targets commercial AI systems capable of listening, speaking, and interacting with external tools.
ET Most Innovative AI Product Awards 2026
The ET Most Innovative AI Product Awards 2026 is inviting nominations for the Agentic AI & Autonomous Systems category. The awards recognize scalable AI innovations driving autonomous decision-making, intelligent workflows, and the future of enterprise technology.
ServiceNow Utilizes AI for Workflow Efficiency
ServiceNow is using its own AI platform, Now on Now, to achieve significant ROI and demonstrate how C-suites can move beyond AI experimentation into tangible, scalable automation. The company is leveraging generative AI and machine learning to improve workflow efficiency and enterprise productivity.
Google Introduces Open Source Agent Executor
Google has introduced an open-source Agent Executor to support AI agents in production. The new runtime could help enterprises address operational challenges tied to scaling AI agents. It includes features such as durable execution, secure sandboxing, and connection recovery.
C-Suite Embraces Shadow AI
Senior decision-makers are the heaviest users of shadow AI in the workplace. A recent study found that 65% of decision-makers use shadow AI, compared with 31% of employees below decision-maker level. Employees continue using AI because they see practical value in it.
37% of Hourly Workers Encounter AI on the Job
A growing share of hourly workers in the United States are encountering artificial intelligence on the job before they feel financially prepared for it. The findings suggest that AI's impact is moving beyond Silicon Valley and corporate offices into various jobs.
ECB Urges Banks to Strengthen Cyber Security
The European Central Bank (ECB) has summoned banks to a meeting to strengthen cyber security systems after new AI models have exposed weaknesses. The ECB will warn banks about risks linked to advanced AI models and urge US banks to share lessons with European rivals.
Sources
- What to know about the AI models that are jolting Washington
- Anthropic’s Claude Mythos AI Uncovers More Than 10,000 Critical Software Vulnerabilities
- NZ at wild frontier of AI superhacking
- Pope Leo calls for ‘disarming’ of artificial intelligence
- Pope Leo warns some AI weapons ‘practically beyond’ human control
- Google showcased Gemini’s focus on practical AI, but I still have mixed feelings
- AI voice bots hijacked by ‘hidden’ sounds in podcasts, MP3 files and YouTube clips
- ET Most Innovative AI Product Awards 2026: Recognise leaders in Agentic AI & Autonomous Systems
- Artificial Intelligence at Service Now - Emerj Artificial Intelligence Research
- Google adds open source Agent Executor to support AI agents in production
- Turns out the C-suite loves shadow AI
- 37% of Hourly Workers Already See AI on the Job
- NEWS: ECB calls banks to urgent AI cyber security meeting
Comments
Please log in to post a comment.