Fireblocks said attackers impersonated its recruiters on LinkedIn and Google Meet, then used GitHub coding tasks to infect candidates with malware in a campaign resembling the Contagious Interview pattern linked to Lazarus Group.
Security researcher Doyeon Park disclosed a critical 0-day vulnerability in Cosmos' CometBFT consensus layer that can deadlock nodes. He slammed officials for ignoring proof-of-concept, downgrading severity to dodge bounty, and marking his report as spam.
The open-source FLARE-AI platform is now live, giving the public a place to report and track AI-related harms. Its launch comes as US lawmakers propose a bill directing NIST to create reporting standards and a centralized AI flaw database.
Japan-based AI startup Sakana AI said on July 21 that it has added a cybersecurity-focused version, Fugu-Cyber, to its Sakana Fugu model family and is offering it through a separate API endpoint. In its announcement, the company said Fugu-Cyber reached an 86.9% success rate on CyberGym, a benchmark for vulnerability validation, and scored 72.1% on CTI-REALM, which measures the ability to convert threat intelligence reports into deployable detection rules. Sakana AI described those results as comparable to specialized frontier security models including OpenAI’s GPT-5.5-Cyber and Anthropic’s Mythos-Preview. The company did not say the model outperformed those systems, and the announcement did not include a ranking table. Sakana AI also stressed that Fugu-Cyber is not a newly trained foundation model. Instead, it is a cybersecurity-specific configuration built on top of the Sakana Fugu orchestration system, which routes tasks across multiple models assigned to Thinker, Worker, and Verifier roles. Access is gated through an application process, with users required to state their intended use and provide verified contact details for manual review. Pricing starts at $6 per million input tokens and $36 per million output tokens, with a higher tier applied once context exceeds 272K tokens.
