ARGUS is a self-hosted, open-source security operations platform designed to bring infrastructure monitoring, threat detection, AI-assisted investigation, and response into one system. It collects signals from cloud services, identity systems, GitHub, Kubernetes, endpoints, and networks. Read-only AI specialists then connect evidence to investigate possible threats, while policy controls govern whether any response can proceed. The platform emphasizes that autonomy is disabled by default: actions require defined scope and authorization, can be checked in dry runs, and are designed to be reversible and recorded in a signed audit trail.
The platform lists seven investigation agents, 27 typed response actions, 13 built-in detection rules, and eight sandbox attack scenarios. Its response safeguards include Slack approvals, limits, kill switches, and rollback. Additional capabilities cover AI governance, including model access, spending budgets, data-loss prevention, and discovery of unauthorized AI use, as well as continuous security testing through authorized penetration testing, agent canaries, and sandbox attacks. ARGUS is offered under the MIT license and can be deployed with Docker Compose, Kubernetes, AWS, or Azure. It supports OpenAI, Anthropic, and Ollama, as well as operation without a model. Its central pitch is security automation with human control, auditability, and no dependence on a single vendor.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Attackers attempted to exploit critical GitLab vulnerability CVE-2026-97229 on Alpine Linux’s instance, crashing worker processes, but Alpine found no evidence of successful access or tampering. Alpine rebuilt the instance, reset passwords, and warns that self-managed GitLab deployments also need patched Ruby - not just a GitLab upgrade. (alpinelinux.org)
An iPhone backup audit reportedly uncovered more than 145,000 deleted photos and videos in a thumbnail cache, including sensitive images. The claim that apps with file access could read them has not been independently verified. (twitter.com)
A leak of DHS records exposed personal and workplace details for more than 41,000 employees and contractors, including undercover ICE agents and information about surveillance programs. Distributed Denial of Secrets published the data, while Okta said it had no evidence its systems were breached. (wired.com)
Several PDF tools generate advertised AES-256 keys from weak randomness, making encrypted files vulnerable to passwordless decryption. MuPDF versions before 1.28.5 expose only 9 bits of random state in the key, while Foxit relies on predictable time-based randomness. (v12.sh)
Distributed Denial of Secrets published a searchable database with contact information for nearly 40,000 DHS employees, reportedly obtained through an Okta single sign-on leak. The group says the database highlights officials overseeing agency policies and operations, including immigration enforcement and covert programs. (techdirt.com)
Distributed Denial of Secrets released personal and job data on more than 40,000 DHS employees, including staff in surveillance, intelligence, and undercover roles. The group says the data came from a hack of DHS’s Okta sign-on service, but Reuters could not verify the full trove and Okta reported no evidence of a compromise. (reuters.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.