A fact-focused update for security and risk professionals, covering roughly the last 24 hours of AI-related security developments.

🔐 Core Security Intelligence

1) 7AI raises $130M to scale AI-agent SOC operations

What’s new
Cybersecurity startup 7AI has raised $130 million in a Series A round, an unusually large amount for early-stage funding. The company builds specialized AI agents that triage, investigate, and in some cases autonomously respond to security alerts in SOCs, with the goal of reducing analyst overload and improving time-to-contain. The round was led by Index Ventures with participation from Greylock, CRV, Spark Capital, and Blackstone’s innovation arm, valuing 7AI at around $700 million.

Source: Cybersecurity Startup 7AI Raises $130 Million in Series A Funding

Why it matters
This is a strong signal that “AI agents for SOCs” are moving from hype to mainstream. If investors are pouring nine figures into an AI-native SOC platform, boards and executives will soon expect similar automation capabilities from their own security programs. At the same time, outsourcing triage and response to opaque AI agents without tight guardrails carries obvious operational, safety, and liability risks.

Defenses (and opportunities)

• Define an “AI-in-the-SOC” policy before buying tools. Clarify what decisions AI agents may take autonomously (e.g., enrichment, ticket updates) versus actions that require human approval (e.g., host isolation, blocking identity providers). This prevents vendors from setting your risk tolerance by default.
• Instrument AI agents like any other privileged system. Ensure every autonomous action is logged, attributable, and reviewable. Require rollback paths and “big red button” kill switches so humans can halt agent behavior quickly if something goes wrong.
• Run controlled pilots with clear success metrics. Start in constrained domains (e.g., low-risk alert categories) and measure MTTR, false-positive handling, and failure modes before scaling to business-critical environments.

Expert insight
AI agents are not just another analytics feature; they are a new class of semi-autonomous operator in your SOC. Treat them more like new staff with privileged access than like a dashboard widget. The organizations that win will be those that combine AI-driven scale with disciplined guardrails, not those that blindly “let the bots run the shop.”

2) Chinese-backed hackers seen using AI to automate campaigns

What’s new
An interview on NPR’s Here & Now with Steven Adair (Volexity) describes how Chinese-backed hacking groups are using AI to automate portions of their operations. According to Volexity’s research, AI is being used to scale phishing, improve lure quality, and assist with aspects of intrusion workflows, enabling campaigns that are more continuous and adaptive than purely manual efforts.

Source: How AI is changing hacking

Why it matters
This confirms that nation-state actors are not just experimenting with AI but operationalizing it to enhance classic tradecraft. AI lowers the marginal cost of each new lure, document, or campaign variant, and can help attackers iterate more quickly when defenses adapt. For defenders, this means more frequent, higher-quality attempts that look increasingly legitimate to both people and filters.

Defenses

• Raise the bar on phishing and social-engineering controls. Move beyond basic secure email gateways to layered defenses: behavioral anomaly detection, robust DMARC/SPF/DKIM enforcement, and strong identity controls that limit blast radius when phishing succeeds.
• Make user training reflect AI-quality content. Traditional “obvious scam” examples are no longer representative; training material should include AI-generated lures that are grammatically correct, context-aware, and personalized.
• Instrument identity and session security as your real perimeter. Assume some AI-enhanced phishing will work. Emphasize phishing-resistant MFA, device and network posture checks, and fine-grained access policies so a stolen credential does not equal full compromise.

Expert insight
AI does not fundamentally change the attacker’s goals, but it dramatically improves their ability to iterate and scale. The right defensive response is not panic; it is a renewed focus on identity, segmentation, and detection depth, assuming that first-layer phishing prevention will fail more often.

3) New study: AI companies’ safety practices fall short of global standards

What’s new
A study coordinated by the Future of Life Institute and reported by Reuters finds that major U.S. AI companies’ safety practices do not meet key global benchmarks for managing advanced AI risks. The report highlights gaps in transparency, incident reporting, independent oversight, and binding safety standards. It argues that despite rising concern about AI-powered hacking, disinformation, and safety failures, AI firms remain less tightly regulated than many other high-risk industries.

Source: AI companies’ safety practices fail to meet global standards, study shows

Why it matters
For security and risk teams, this is not an abstract governance debate. If upstream AI providers are under-regulated and under-tested, enterprises that build on their APIs and models inherit systemic risk they cannot fully observe or control. Regulatory pressure will likely shift toward customers as well, with expectations for due diligence, third-party risk management, and AI-specific controls.

Defenses

• Treat AI providers as critical third parties, not generic SaaS. Include AI vendors in your vendor-risk program, demand detailed security and safety documentation, and push for clear SLAs around incident notification and misuse handling.
• Document your dependency on specific models and APIs. Maintain an inventory that maps business processes to the AI services they rely on, so you can respond quickly if a provider experiences a safety failure or regulatory restriction.
• Design for portability wherever possible. Minimize tight coupling to a single model or vendor. Use abstraction layers so you can swap models or providers if safety posture or regulatory status changes.

Expert insight
Many organizations are moving faster on AI adoption than regulators or standards bodies can keep up. That puts the burden on you to create internal “mini-regulation” through contracts, architecture, and policy. The more critical AI becomes to your operations, the more you should think like a regulator evaluating a systemic risk, not just a buyer shopping for features.

4) “5 threats that reshaped web security” highlights AI-driven supply-chain attacks

What’s new
A new feature on The Hacker News reviews five major web-security shifts in 2025, including a section on AI-powered software supply chain attacks. The article notes a 156% surge in malicious packages on open-source repositories and describes the “Shai-Hulud Worm,” a self-replicating campaign that used AI-generated scripts and AI-aware evasion to compromise over 500 npm packages and 25,000+ GitHub repositories. The worm leveraged AI tooling for reconnaissance and was explicitly designed to trick both traditional security tools and AI-based scanners.

Source: 5 Threats That Reshaped Web Security This Year [2025]

Why it matters
This is a consolidated view of how AI is already embedded in the offensive toolchain around web and app security. Instead of a hypothetical future, it describes real campaigns where AI was used to generate polymorphic malware, evade AI-powered detectors, and weaponize CI/CD pipelines for mass distribution. For organizations relying heavily on JavaScript, npm, and CI-based deployments, this is a warning about how fragile the current supply chain can be under AI pressure.

Defenses

• Shift to behavioral and provenance-based validation for dependencies. Do not rely solely on source reputation or static allowlists. Track where packages come from, how they change, and what they do at runtime, especially in build pipelines.
• Treat AI-generated code and scripts as untrusted input. Require security review, secrets scanning, and testing for LLM-produced code just as you would for junior developer contributions, and codify that in your SDLC policy.
• Implement strong CI/CD hardening. Lock down tokens, enforce least privilege for build agents, monitor for anomalous pre/post-install hooks, and ensure that publishing flows are protected against credential theft campaigns like those highlighted in the article.

Expert insight
The pattern here is clear: attackers are applying AI wherever developers do, especially in the supply chain. Security leaders should treat software-supply-chain resilience and AI-aware detection as a single, unified priority rather than two separate projects.

📌 Additional Watch: Very Recent AI-Security Story (Past ~48 Hours)

These items fall just outside a strict 24-hour window but are important enough to keep on your radar.

Malicious npm package uses embedded prompt to evade AI security tools

What’s new
Researchers disclosed a malicious npm package, eslint-plugin-unicorn-ts-2, that combines classic supply-chain tricks (typosquatting, post-install credential theft) with an embedded prompt aimed at influencing AI-driven security scanners. The package includes text telling AI tools that the code is legitimate and tested, while a post-install script quietly exfiltrates environment variables (including API keys and tokens) to a remote webhook.

Source: Malicious npm Package Uses Hidden Prompt and Script to Evade AI Security Tools

Why it matters
This is one of the clearest cases so far of attackers intentionally targeting AI-based security tooling, not just human reviewers. It demonstrates that as defenders rely more on AI to triage and score code, attackers will experiment with “prompt-level” obfuscation and manipulation to downgrade or misclassify malicious artifacts.

📊 At-a-Glance Summary

Categories: Artificial Intelligence

Tags: AI Security, AI Supply Chain, Model Safety, Nation-State Threats, Prompt Injection, SOC Automation