A concise, fact-based update for security and risk professionals. Core technical stories first, then broader context.

🔐 Core Security Intelligence

1) Oracle rushes emergency patch for Cl0p-exploited zero-day in EBS

What’s new:
Over the weekend, Oracle released an emergency patch for a critical zero-day vulnerability in its E-Business Suite (EBS) that the Cl0p ransomware group has actively exploited. The flaw is tracked as CVE-2025-61882.
Source: SecurityWeek
Also covered by: The Register

Why it matters:
This is a stark reminder that enterprise ERP systems remain high-value targets. Attackers are not waiting for monthly patch cycles—they’re exploiting fresh holes immediately. Organizations running Oracle EBS must prioritize patch deployment and validate integrity post-update.

Defenses:

• Deploy Oracle’s emergency patch immediately. Delaying leaves systems exposed to active exploit attempts, especially in environments where attackers scout for just-in-time vulnerabilities.
• Audit related modules and custom extensions. Even after patching, custom code or plugins might reintroduce weaker paths; ensure they’re tested against the updated baseline.
• Enhance monitoring and anomaly detection. Watch for suspicious data access, unauthorized privilege escalations, or lateral movement in the same window as patch rollout.

Expert Insight:
This incident shows how the “patch window” is shrinking—attacks follow days or hours, not weeks. ERP systems like EBS are too frequently neglected but are often deeply integrated with business logic and data. Defenders must treat those systems with urgency and verify that downstream components remain secure after patch application.

2) CometJacking: Single URL hijacks agentic browser to exfiltrate data

What’s new:
Researchers disclosed CometJacking, an exploit against Perplexity’s AI browser (Comet). A crafted URL can trigger hidden prompts that exfiltrate data from connected services (email, calendar, memory) without user awareness.
Source: The Hacker News

Why it matters:
Agentic AI browsers blur the boundary between browsing and tool access. This exploit demonstrates how seemingly innocuous navigation can hijack trusted agent memory and data flows. Any browser-based AI system must assume prompt takeover is possible.

Defenses:

• Sanitize and validate all URLs and prompt inputs. Treat any link entering the agent’s context with skepticism; block or sandbox suspicious payloads.
• Limit agent permissions to memory and connected services. Use least privilege in what the AI browser can access (e.g. email, calendar) to reduce blast radius.
• Monitor agent output or exfiltration signals. Baseline agent behavior and flag any unexpected data flows or memory dumps immediately.

Expert Insight:
CometJacking is a clear example of the emerging attack class: malicious prompts inside trusted agents. The exploit bypasses normal UI controls because the agent acts autonomously. Defenders need to design agents assuming they will be targeted—not just their inputs. Control, validation, and observability in agent systems become non-optional.

3) CISOs rethink security organizations for the AI era

What’s new:
CSO Online published a piece arguing that as AI introduces new risks—deepfakes, agent abuse, inadvertent IP leaks—security org charts must evolve. The author suggests structuring teams around AI risk domains, not traditional siloed functions.
Source: CSO Online

Why it matters:
Traditional security models were built around perimeter, identity, endpoint. AI changes the attack surface. Teams will need new roles, responsibilities, and lines of communication to catch AI-native threats. This is a signal that organizational design matters as much as tooling.

Defenses:

• Define AI risk domains explicitly. Create roles or functions (e.g. Agent Risk, Model Trust, Prompt Security) within your org so responsibilities are clear.
• Incorporate AI context into every review and audit. Ensure that regular audits, threat models, and red teaming include AI-specific vectors (e.g. prompt injection, memory leakage).
• Foster alignment across teams (data, ML, security). AI risk is cross-domain—security should collaborate early in ML development, not as an afterthought.

Expert Insight:
Organizational inertia is one of the biggest barriers to AI security maturity. Even with top tools, if no team owns prompt trust or agent boundaries, gaps persist. Designing around risk domains ensures visibility, responsibility, and better cross-collaboration as AI scales inside your enterprise.

🌐 Extended Reading / Broader AI Risk & Governance

4) California’s AI safety law takes effect

What’s new:
California’s recently signed bill (SB 53) now mandates that high-compute AI systems publicly disclose safety measures, report safety incidents, and establish whistleblower protections.
Source: AP News
Also covered: Le Monde

Why it matters:
This may set a precedent for state-level AI regulation in the U.S. Companies deploying compute-heavy models or offering agent systems must prepare for disclosure and compliance obligations.

5) AI-powered vulnerability detection may outstrip patch capacity

What’s new:
A report warns that AI-based vulnerability discovery is accelerating faster than organizations can fix them, risking a backlog of unaddressed security flaws.
Source: Cybersecurity Dive

Why it matters:
When detection outpaces remediation, risk accumulates. Organizations must balance the flood of findings with prioritization and realistic capacity, not just alert volume.

⚠️ Updates / Follow-ups

No significant updates beyond weekend patches and exploit disclosures.

Summary Table

Categories: Cybersecurity News

Tags: Agentic Risk, AI Regulation, AI Security, Org Structure