A concise, fact-based update for security and risk professionals. Core technical stories first, followed by broader AI risk & policy developments.

🔐 Core Security Intelligence

1) AI browser “sidebar spoofing” attack demonstrated

What’s new:
Researchers at SquareX Labs demonstrated a proof-of-concept attack called “sidebar spoofing,” which targets AI-enabled browser sidebars (such as those in Perplexity Comet or Microsoft Edge’s Copilot Mode). A malicious extension with host/storage permissions injects JavaScript to display a fake sidebar that appears legitimate but routes user prompts through attacker-controlled LLMs.
Source: SC Magazine

Why it matters:
AI-embedded browsing changes the trusted UI paradigm. When a malicious extension can hijack the AI assistant’s prompt channel, it opens new vectors for data exfiltration, phishing, or stealthy payloads disguised as legitimate AI responses.

Defenses:

• Restrict extension permissions and audit metadata. Only allow browser extensions with minimal permission (avoid “host & storage” where possible) and review first-party reviews or telemetry for unusual behaviors.
• Validate prompt channels and UI integrity. Monitor for UI overlays or unexpected prompt destinations; implement UI whitelisting or trusted context checks before routing queries to LLMs.
• Educate users about trusted integration surfaces. Highlight that AI sidebars can be spoofed; encourage prompt routing only through verified toolbar icons or built-in features, not unfamiliar extensions.

Expert Insight:
This attack reminds us that adding AI to familiar interfaces doesn’t reduce risk, it reframes it. Browsers were once a well-understood frontline; now they’re agentic platforms. Defenders must treat browser extensions as part of the AI supply chain and assume attackers will aim at UI hijacks and prompt redirection.

2) Microsoft 2025 threat report: AI-driven cyberattacks blur identity, cloud and ransomware boundaries

What’s new:
Microsoft’s latest report highlights accelerating trends: identity-based attacks rose 32% in H1 2025, ransomware in the cloud inflated to 40% of cases, and generative-AI tools now aid lateral movement, exploit discovery, and real-time control evasion.
Source: Help Net Security

Why it matters:
The convergence of identity, cloud, and AI means traditional siloed protections no longer suffice. Attackers exploit gaps between identity controls, cloud misconfigurations, and AI-tool workflows to launch hybrid assaults.

Defenses:

• Treat identity as the primary asset. Secure non-human identities (service accounts, scripts, bots) with MFA, conditional access policies, and audit privilege escalation paths.
• Harden cloud-native controls. Monitor cloud API usage, enforce least-privilege identity roles, and segment workloads to reduce lateral movement from compromised identities.
• Integrate AI-aware detection. Use anomaly detection tuned to AI-driven behaviors (rapid prompt bursts, automated recon, model-tool chaining) and incorporate them into your SOC processes.

Expert Insight:
Microsoft’s framing signals a paradigm shift: it’s not just “cybersecurity” versus “AI”. It’s identity-centric, cloud-native, AI-enabled adversaries. Security teams must evolve from reactive patching to proactive identity and behavior-based defense—because speed, scale, and automation now reside with the attacker.

🌐 Extended Reading / Broader AI Risk & Governance

3) Nearly half of organizations impacted by AI security flaws (EY report)

What’s new:
Ernst & Young (EY) found that 50% of companies surveyed reported negative impacts from vulnerabilities in their AI systems, while only 14% of CEOs believe their AI assets adequately protect sensitive data.
Source: Cybersecurity Dive

Why it matters:
While AI adoption accelerates, maturity lags. Insecurity in AI systems means new attack vectors—not just on endpoints or networks, but inside the models and data flows themselves.

⚠️ Updates / Follow-ups

No major updates today that meet the criteria for follow-up on previously covered stories.

Summary Table

Categories: Cybersecurity News

Tags: AI Security, AI Vulnerabilities, Browser Security, Cloud Risk, Identity Threats