OpenAI has unveiled Daybreak, its answer to Anthropic’s Claude Mythos, amid a growing market for frontier AI-powered cyber defense platforms. The initiative combines OpenAI’s large language models, Codex’s agentic capabilities, and integrations with the broader enterprise security ecosystem.

The company said Daybreak is focused on accelerating cyber defense operations and enabling organizations…

The US Cybersecurity and Infrastructure Security Agency (CISA) does not yet have access to Anthropic’s bug-hunting AI model, Claude Mythos, even though other government agencies do, Axios reported earlier this week.

As if that weren’t a big enough slap in the face for the national cyber-defense agency, the list of those who do have access to Mythos includes several unauthorized users, according…

The US Cybersecurity and Infrastructure Security Agency (CISA) does not yet have access to Anthropic’s bug-hunting AI model, Claude Mythos, even though other government agencies do, Axios reported earlier this week.

As if that weren’t a big enough slap in the face for the national cyber-defense agency, the list of those who do have access to Mythos includes several unauthorized users, according…

European regulators have largely been frozen out of early access to Anthropic’s new Mythos model, Politico reports. The AI technology, aimed at cybersecurity use cases, is said to be able to identify and exploit technical vulnerabilities at a level that surpasses most humans — signaling a structural shift for CISOs and the cybersecurity industry.

For security reasons, Anthropic has chosen to…

Move over shadow IT; shadow AI is the new risk on the scene. The explosion of available AI tools, leadership’s enthusiasm for the new technology, the push for employees to do more with less, nascent governance and the sheer speed at which AI is evolving has created the perfect environment for shadow AI to flourish.

“Every CISO I talk to has discovered some form of shadow AI,” says Andrew Walls,…

Security researchers have uncovered six high-to-critical flaws affecting the open-source AI agent framework OpenClaw, popularly known as a “social media for AI agents.” The flaws were discovered by Endor Labs as its researchers ran the platform through an AI-driven static application security testing (SAST) engine designed to follow how data actually moves through the agentic AI software.

The…

Enterprise security teams racing to enable generative AI tools may be overlooking a new risk: attackers can abuse web-based AI assistants such as Grok and Microsoft Copilot to quietly relay malware communications through domains that are often exempt from deeper inspection.

The technique, outlined by Check Point Research (CPR), exploits the web-browsing and URL-fetch capabilities of these…

MS 연구에 따르면, 겉보기에는 무해해 보이는 단 하나의 프롬프트만으로도 주요 언어 및 이미지 모델의 안전 가드레일을 체계적으로 제거할 수 있는 것으로 나타났다. 이는 기업 환경에 맞춰 모델을 맞춤화하는 과정에서 AI 정렬의 지속 가능성에 대한 새로운 의문을 제기한다.

MS 연구진은 공식 블로그를 통해 ‘GRP-오블리터레이션(GRP-Obliteration)’이라 명명한 이 기법이, 본래 모델을 더 유용하고 안전하게 만들기 위해 활용되는 일반적인 AI 학습 방식인 그룹 상대 정책 최적화(Group Relative Policy Optimization)를 역으로 활용해 정반대의 효과를 낸다고 설명했다.

MS는 GPT-OSS, 딥시크 R1 디스틸(DeepSeek-R1-Distill) 계열, 구글의 젬마,…

A single benign-sounding prompt can systematically strip safety guardrails from major language and image models, raising fresh questions about the durability of AI alignment when models are customized for enterprise use, according to Microsoft research.

The technique, dubbed GRP-Obliteration, weaponizes a common AI training method called Group Relative Policy Optimization, normally used to make…

This year will mark the turning point where artificial intelligence will stop _assisting_ and start _acting._ We will witness a qualitative leap towards agent-based or agentive AI, capable of making autonomous decisions, managing complex workflows, and executing end-to-end tasks without constant intervention. However, this autonomy carries with it a serious warning for businesses: the ability to…