A concise, fact-based update for security and risk professionals. Past 24 hours only. We exclude prior stories unless there is a material update.
🔐 Core Security Intelligence
1) New backdoor “SesameOp” abuses OpenAI’s API for covert C2
What’s new:
Microsoft researchers detailed SesameOp, a stealth backdoor that uses the OpenAI Assistants API as a command-and-control channel, blending attacker traffic with normal model requests. The malware retrieves prompts as “tasks” and exfiltrates results through the same API path, complicating detection.
Source: Microsoft detects SesameOp backdoor leveraging OpenAI API
Why it matters:
Using legitimate AI APIs for C2 raises the bar for defenders. Network filters that allow sanctioned AI services can become blind spots, and telemetry often treats model calls as trusted activity. This is an early example of “living-off-the-AI-land” tradecraft that will likely spread.
Defenses:
- Segment and monitor AI egress. Create distinct egress paths for AI APIs with separate logging and anomaly baselines.
- Bind AI access to identities and hosts. Use per-host tokens and flag model calls from unexpected binaries.
- Correlate AI calls with host telemetry. Combine API, EDR, and process data to detect misuse patterns.
Expert Insight:
Attackers will increasingly hide inside the same AI workflows you adopt for productivity. Treat AI API usage like any other privileged integration with identity binding, audit logs, and behavioral analytics.
2) Android November update patches critical RCE in System component
What’s new:
Google’s November 2025 Android security update fixes multiple System flaws, including a critical remote code execution vulnerability exploitable without user interaction.
Source: Android update patches critical remote code execution flaw
Why it matters:
Mobile devices now hold sensitive AI session data and enterprise tokens. A System-level RCE provides attackers with privileged footholds on AI-enabled endpoints.
Defenses:
- Enforce patch compliance via MDM. Push updates immediately and monitor coverage for privileged users.
- Add runtime protections. Mobile EDR and network segmentation can detect lateral movement while updates propagate.
- Restrict risky configurations. Disable sideloading and limit AI assistant apps to trusted sources.
Expert Insight:
Mobile RCEs rapidly become weaponized. The mix of stored credentials and AI app tokens makes prompt patching non-negotiable.
3) Zscaler acquires SPLX to extend AI security across the lifecycle
What’s new:
Zscaler acquired SPLX, an AI-security startup specializing in model discovery, automated red teaming, and governance tied to the Model Context Protocol.
Source: Zscaler acquires SPLX to strengthen AI security lifecycle coverage
Why it matters:
Security vendors are consolidating around end-to-end AI governance. Enterprises want unified discovery, testing, and enforcement for models and agents.
Defenses:
- Map your AI assets. Identify all models, datasets, and tools under development.
- Embed adversarial testing. Run automated jailbreak and poisoning tests in CI/CD.
- Close the compliance loop. Integrate AI governance policies with runtime enforcement.
Expert Insight:
AI security is becoming a lifecycle problem. Integrating discovery and governance at design time will separate resilient programs from reactive ones.
4) Cisco introduces secure, automated networking for distributed AI
What’s new:
Cisco unveiled new networking innovations aimed at AI workloads, adding built-in segmentation, automation, and telemetry across data-center, campus, and industrial environments.
Source: Cisco unveils secure network innovations for AI-driven environments
Why it matters:
AI traffic now dominates enterprise backbones. Embedding security directly into switching and routing helps prevent lateral exposure of model and data pipelines.
Defenses:
- Segment AI data flows. Isolate training and inference traffic from general network paths.
- Enable flow telemetry. Alert on spikes, protocol changes, or unauthorized AI destinations.
- Secure the edge. Authenticate industrial AI sensors and enforce encrypted transport.
Expert Insight:
Securing AI is as much a network problem as an application one. Data-centric segmentation is the new perimeter.
5) Graylog adds AI-driven investigation and MCP server access for SOCs
What’s new:
Graylog 7.0 introduces AI-driven investigation features, Model Context Protocol (MCP) server connectivity, and AWS Security Data Lake integration to accelerate response.
Source: Graylog 7.0 introduces AI features and MCP integration
Why it matters:
SOC integrations with AI agents and MCP servers can boost speed but create new trust boundaries. Proper identity and audit controls are essential.
Defenses:
- Gate agent actions. Require approvals for AI-driven containment or enrichment tasks.
- Lock down MCP endpoints. Rotate credentials and restrict available tools.
- Preserve evidence chains. Tag AI outputs with provenance and audit context.
Expert Insight:
AI-assisted SOCs are powerful but risky. Guardrails and visibility must evolve alongside automation.
6) Most websites remain unprepared for AI agent traffic
What’s new:
A new analysis finds that only 2.8% of sites have protections capable of identifying or filtering AI-agent traffic, warning that legacy bot defenses are inadequate.
Source: Most websites aren’t ready for AI-agent traffic
Why it matters:
Agentic browsing, scraping, and transaction automation are redefining “bot” behavior. Websites built for human traffic will face new data-exfiltration and abuse patterns.
Defenses:
- Adopt intent-based detection. Move beyond user-agent filtering to contextual risk scoring.
- Monitor API usage anomalies. Detect machine-generated access across account sessions.
- Update abuse playbooks. Include AI-agent activity in fraud and DDoS response plans.
Expert Insight:
AI agents are the new web clients. Visibility into their purpose and intent will be the next frontier of web defense.
⚠️ Updates / Follow-ups
No major updates to previously covered stories within the past 24 hours.
Summary Table
| Topic | Key Risk | Defense Highlights |
|---|---|---|
| SesameOp backdoor | Abuse of legitimate AI APIs for C2 | Segment AI egress; bind tokens; correlate host telemetry |
| Android RCE | Mobile fleet compromise | Rapid patching; mobile EDR; restrict sideloading |
| Zscaler + SPLX | AI lifecycle blind spots | Inventory models; automate red-team tests; unify governance |
| Cisco AI networking | Lateral exposure via AI data flows | Microsegment traffic; enable telemetry; secure edge devices |
| Graylog 7.0 | SOC automation trust issues | Gate agent actions; protect MCP; maintain evidence integrity |
| Website AI-agent surge | Inadequate defenses vs agentic abuse | Intent-based detection; monitor APIs; expand fraud playbooks |
Categories: Cybersecurity News
The SesameOp case is the one worth watching closely. “Living-off-the-AI-land” is a natural evolution of living-off-the-land attacks — same logic, harder to detect because the traffic is legitimately sanctioned. Most security teams haven’t built behavioral baselines for AI API usage yet, which is exactly the gap being exploited.