As concerns spread over cyber intrusions linked to using AI models to discover software vulnerabilities, industry moves to seek responses through cooperation are gaining momentum. After Anthropic introduced its Project Glasswing programme in April to help major companies and institutions find and respond to software vulnerabilities with its AI model Mythos, OpenAI rolled out the AI-based cybersecurity initiative Daybreak to speed up vulnerability identification, patch verification and security hardening.
Palo Alto Networks introduced the Frontier AI Defense initiative to respond to cyber threats using advanced AI models. Frontier AI Defense focuses on combining an AI-native security platform, Unit 42 consulting and threat expertise, and strategic partners to respond more effectively to security threats involving AI.
• OpenAI to launch AI-based cybersecurity initiative Daybreak • Palo Alto Networks: Mythos found vulnerabilities in 3 weeks at a level comparable to a year of manual penetration testing
Experts are also sharing experiences using Anthropic Mythos and OpenAI's GPT-5.5 cyber model to identify software vulnerabilities.
Palo Alto Networks said it found 75 vulnerabilities in its products within a month after it began using high-performance AI cybersecurity models from Anthropic and OpenAI. That is more than 7 times its monthly average of 5 to 10 findings.
• Palo Alto Networks tests Mythos and GPT-5.5-Cyber to find vulnerabilities in its products • Palo Alto Networks: Mythos found vulnerabilities in 3 weeks at a level comparable to a year of manual penetration testing
Some have also responded that the Anthropic Mythos AI model's ability to find software vulnerabilities is not as strong as it is known to be.
• Mythos finds only 1 low-risk vulnerability in open-source tool curl...Anthropic exaggerated marketing?
Cases that raise concern about attacks using AI models also continue to emerge.
Google said it confirmed the first real-world case in which cyber criminals used AI to discover and weaponise a zero-day vulnerability. As Anthropic's AI model Mythos finds thousands of software vulnerabilities and banks, technology companies and governments around the world are on edge, security experts said it is already possible to reproduce such capabilities even with existing models.
• Google spots first AI exploit abusing zero-day vulnerability in the field • Security experts on Mythos shock: Zero-day vulnerability detection is already possible with existing AI models
Moves by countries to secure access to Anthropic Mythos and OpenAI's GPT-5.5 cyber model are also accelerating. The South Korean government is also active. The Ministry of Science and ICT is actively sounding out participation in Anthropic's cybersecurity consultative body Project Glasswing. The government will also announce a comprehensive response plan for AI-driven cybersecurity threats as early as late May.
• Can the Ministry of Science and ICT gain access to Anthropic Mythos? • "Opus 4.7 is also a sufficient threat"...government to unveil Mythos response plan as early as late May • Ministry of Science and ICT discusses AI security responses such as Mythos with industry, academia and research
In the European Union, discussions are under way on access to OpenAI's GPT-5-5 cyber security model, while Anthropic is not disclosing Mythos to the EU. Reports said Japan's three megabanks are expected to secure access to Anthropic's AI model Mythos soon.
• EU discusses GPT-5-5 cyber access with OpenAI...Anthropic's Mythos still not available • "Japan's three megabanks to secure access to Anthropic Mythos within 2 weeks"
The International Monetary Fund urged that AI models such as Anthropic's Mythos be treated as a systemic risk to the overall financial system, not an operational issue for individual financial firms.
• New cyber threat revealed by Mythos...IMF: The entire financial system could be shaken
Other moves by companies at home and abroad around security are summarised below.
Palo Alto Networks introduced the identity security platform Idira, which integrates and manages identities for people, machines and AI agents under a single privileged access management framework. Cisco released the Foundry Security Spec as open source to standardise security assessment of IT systems using large language models.
• Palo Alto Networks launches Idira platform integrating identities of people, machines and AI agents • Cisco releases open-source framework to standardise LLM-based security assessment
Illumio, a global breach containment specialist, stressed the risks of lateral movement, an emerging challenge in corporate cybersecurity. It said organisations should secure visibility into attacker movement paths and make segmentation, or microsegmentation, central to a zero trust security strategy to prevent the spread of breaches.
• "Security focused on prevention and detection is over...Zero trust requires an era of containment"
Google is introducing Intrusion Logging on Android, a security feature that records and analyses traces of spyware compromise.
• Google rolls out Android intrusion logging...strengthens spyware investigation features
NH Nonghyup Bank signed a memorandum of understanding with the Financial Security Institute to verify digital asset service technology and strengthen security.
• NH Nonghyup Bank strengthens digital asset security cooperation with Financial Security Institute
From the second half of this year, about 1,700 high-risk systems, including major public systems and those processing large amounts of personal information, will undergo regular inspections. Companies and institutions that made proactive investments in personal information protection will also be given meaningful incentives. The Personal Information Protection Commission reported a plan to shift to a prevention-focused personal information management system at a Cabinet meeting chaired by the president.
• Personal information protection system shifts to prevention focus in earnest...punitive fines and incentives in parallel