A fact-based update for security and risk professionals, focused on how AI is reshaping the threat landscape and the defensive stack.

🔐 Core Security Intelligence

1) “Lies-in-the-Loop” attack technique turns AI safety workflows into execution risk

What’s new
Security researchers disclosed a novel attack pattern called Lies-in-the-Loop (LITL) that abuses AI safety and validation workflows themselves. By carefully manipulating inputs and feedback mechanisms, attackers can bypass guardrails and, in some implementations, drive behaviors that escalate into remote code execution risks.

Source:
Lies in the Loop Attacks Transform AI Safety Conversations Into Remote Code Execution Threats

Why it matters
This technique highlights that AI safety controls are not neutral observers. When safety systems influence execution paths, decision logic, or downstream automation, they become part of the attack surface. Organizations relying on layered AI safety checks must consider how those layers interact under adversarial pressure.

Defenses

• Review AI safety pipelines to ensure safety evaluation components cannot trigger execution paths or system state changes.
• Conduct adversarial testing that targets multi-step feedback loops, not just direct prompt injection.
• Isolate safety analysis tooling in restricted execution environments with no access to production resources.

Expert insight
AI safety mechanisms are increasingly complex and interconnected. Treating them as first-class assets in threat modeling is essential to prevent defensive logic from being turned into an exploit primitive.

2) Ciphero raises $2.5M to deliver runtime security controls for enterprise AI

What’s new
AI security startup Ciphero announced a $2.5 million pre-seed funding round to expand its platform for real-time monitoring and verification of enterprise AI usage. The company focuses on enforcing policy constraints, detecting anomalous model behavior, and reducing misuse across large-scale AI deployments.

Source:
Ciphero Raises $2.5M Pre-Seed Round

Why it matters
Most enterprises lack runtime visibility into how AI systems are actually used once deployed. Investment in tools that monitor AI behavior in real time reflects a growing recognition that governance, logging, and enforcement must extend into production environments.

Defenses

• Define enforceable AI usage policies and map them to measurable runtime signals.
• Integrate AI telemetry into existing SOC workflows so anomalous behavior is triaged alongside other threats.
• Validate that monitoring tools capture both inputs and outputs to support forensic review.

Expert insight
AI security is shifting from pre-deployment review to continuous assurance. Runtime controls will likely become a baseline expectation as AI systems take on more autonomous roles.

3) Anubis open-source AI firewall released to counter large-scale bot scraping

What’s new
A new open-source project named Anubis was released to help websites and APIs defend against aggressive scraping by AI-powered bots. The tool uses proof-of-work challenges and behavioral heuristics to increase the cost of automated abuse while preserving access for legitimate users.

Source:
Anubis: Open-Source Web AI Firewall to Protect from Bots

Why it matters
Uncontrolled scraping enables data leakage, competitive intelligence harvesting, and downstream model poisoning. As AI-driven agents scale, traditional rate limiting alone is insufficient to protect sensitive content and APIs.

Defenses

• Deploy challenge-based controls on high-risk endpoints that expose proprietary or sensitive data.
• Monitor scraping behavior for patterns consistent with AI-driven automation rather than human browsing.
• Correlate scraping activity with downstream data usage to identify potential leakage into external models.

Expert insight
Raising the cost of abuse is a practical defense against AI-powered scraping. Active controls such as proof-of-work provide friction that passive detection cannot.

4) Darktrace enhances email defenses to detect AI-driven cross-domain attacks

What’s new
Darktrace announced enhancements to its Darktrace EMAIL platform focused on detecting cross-domain and AI-driven social engineering attacks. The updates emphasize behavior-based analysis across email and collaboration tools to identify subtle anomalies that evade static filtering.

Source:
Darktrace Enhances Email Security to Counter Cross-Domain Attacks

Why it matters
AI-generated phishing and impersonation attacks increasingly blend across email, chat, and collaboration platforms. Defenses that understand business communication patterns are better positioned to detect these multi-channel campaigns.

Defenses

• Use behavior-based detection to establish baselines for legitimate communication patterns.
• Correlate signals across email, identity, and endpoint telemetry to identify coordinated attacks.
• Regularly test defenses with AI-generated social engineering scenarios.

Expert insight
Context-aware detection is critical as AI-driven attacks reduce traditional indicators of compromise. Understanding how people normally communicate is becoming as important as understanding malware signatures.

⚠️ Updates & Follow-ups

No material updates to previously covered AI-security stories were published in the past 24 hours.

📊 At-a-Glance Summary

Categories: Cybersecurity News

Tags: AI Abuse, AI Security, Bot Defense, Email Security, Governance, Prompt Injection