This panel, hosted by Dark Reading, brought together leading experts to explore the dual-edged nature of AI in cybersecurity—how it's empowering defenders and attackers alike.

Key Themes Discussed:

1. AI as a Double-Edged Sword

• Offensive Use by Threat Actors: AI is lowering the barrier for cybercriminals by enabling more personalized, scalable, and sophisticated attacks. Social engineering, phishing, and data exploitation (especially using breached data and deepfakes) are major concerns.
• Quiet Before the Storm: Caleb Sima noted that while AI threats are real, current attacker adoption is in a “quiet period” as they figure out how to scale AI effectively—true productization is expected within a year.

2. Emerging Threats

• Deep Research Tools: Tools that combine breached data, OSINT, image analysis, and LLMs can instantly build rich attacker profiles—dramatically increasing the effectiveness of social engineering.
• AI Agents and Prompt Injections: Rob Ragan highlighted that 2025 will be the year of agentic AI systems. Attackers will exploit these multi-step agent workflows using indirect prompt injection to gain access to sensitive data or execute malicious actions.
• Legacy Infrastructure at Risk: AI is particularly good at analyzing and reverse engineering old, obscure systems, which could lead to an increase in zero-day exploits in critical infrastructure.

3. AI in Defensive Security

• Enhanced Testing and Automation: AI is already improving vulnerability scanning, code review, and prioritization of security issues. Bishop Fox’s team, for example, used AI to expedite firmware vulnerability discovery.
• Code Security: While AI-generated code can increase productivity, it may lack best practices or have hidden flaws. Organizations must double-check LLM-generated code using traditional and AI-enhanced SAST tools.

4. Human Factor and Trust

• Social Engineering Evolution: With convincing deepfakes and AI-generated pretexts, traditional cues for detecting fraud are eroding. Both Eric Kruse and Rob Ragan emphasized technical controls, behavioral analysis, and user education as key defenses.
• Supply Chain and Trust Challenges: AI expands the attack surface across digital supply chains. Knowing who and what to trust is becoming increasingly difficult.

5. Practical Recommendations

• Defense-in-Depth Is Critical: Technical mitigations (e.g., session security, access control for agents, proper validation) are essential.
• Test Models Like Code: Just like traditional code, models will need versioning, testing, and security evaluation. Trust in model integrity and origin will be a new frontier.
• Prepare for AI-Generated Exploits: AI will soon enable junior attackers to find zero-days. Organizations should rethink vulnerability management and threat modeling.

6. Optimism and Opportunities

• Despite concerns, panelists expressed hope. AI can help defenders become faster, more efficient, and creative—especially in threat detection and remediation. Rob Ragan described it as a “superpower” that enables security teams to do more with less.

Panelists:

1. Caleb Sima – Chair, CSA AI Safety Initiative
2. Rob Ragan – Principal Technology Strategist, Bishop Fox
3. Stephen Thoemmes – Developer Advocate, Snyk
4. Erich Kron – Security Awareness Advocate, KnowBe4