Agentic Development Security | Manoj Nair, Snyk | RAISE Summit 2026

Market Scale & Financial Context

• Global software spend is projected at $1.4 trillion this year.
• Coding represents a $1 trillion Total Addressable Market (TAM) for AI, identified as the number one AI use case.
• Snyk serves approximately 5,000 customers, including half of the Fortune 500.
• Recent market sentiment has shifted from token optimization concerns to explicit discussions on cost and ROI.

Quantified Security Risks

• Data from 4,800 customers over the last year reveals a 108% increase in security vulnerabilities directly correlated with the shift to agentic development.
• Remediation efforts are increasing, yet the total volume of vulnerabilities continues to double, indicating a structural compounding of risk rather than a removal.
• Agents introduce new failure modes, including "long-running loops" where agents may "go off the rails" or exhibit unpredictable behavior.

Agent Capabilities & Limitations

• Current benchmarks indicate the best AI models produce only ~60% secure code autonomously, underperforming compared to human expectations.
• Agents accessing MCP servers can exploit "skills" libraries; research found one-third of these libraries contained issues, enabling "toxic flows" and prompt injection attacks that exfiltrate private source code.
• Frontier models currently show 0% failure rates on PII retention attacks but are vulnerable to jailbreaking (30–40% success rate in violating guardrails).
• Conversely, specific open models demonstrated 100% adherence to governance rules, though they currently lag in overall capability compared to frontier models.

Emerging Threat Landscape

• The Five Eyes intelligence agencies estimate autonomous attacks are not months away but are already occurring within weeks to months.
• Attackers no longer need to wait for specific vulnerabilities; with speed gains of up to 100x, a single flaw can be exploited instantly at scale.
• Open models are projected to reach the capability of proprietary models within three to six months, expanding the attack surface significantly.

Strategic Blueprint for Agentic Security

• Eliminate Technical Debt: Organizations are urged to eradicate existing legacy vulnerabilities (e.g., backlog zero) before introducing agents to prevent compounding risks.
• Context-Driven Remediation: Security teams must feed specific context (e.g., breakability checks) to development agents to ensure safe, automated remediation of critical issues.
• Independent Validation: A fundamental architectural shift is required to move from self-validated code to an independent validator, preventing "the fox guarding the henhouse."
• Prevention Over Remediation: Security context must be integrated into the build phase (e.g., within tools like Cursor or Codex) to prevent the creation of vulnerabilities rather than fixing them later.
• Supply Chain Verification: All external skills and MCP servers must be validated before agent access to prevent malicious code execution from third-party libraries.

Regulatory & Operational Implications

• The EU AI Act provides a compliance timeline with enforcement extending to 2027, though immediate architectural shifts are recommended.
• Enterprises are facing board-level pressure to define strategies against autonomous threats, necessitating clear visibility into AI components used in software.
• Snyk has made its remediation tools available to the open-source community, allowing maintainers to fix critical issues without cost.

Call to Action

• Leaders are advised to challenge vendors on their independent validation processes and to reject the reliance on single-source security solutions.
• The industry is encouraged to invest in and support open-source security and trusted open models rather than consolidating around closed frontier labs.
• Organizational structures must evolve from fragile "Jenga tower" models to resilient, load-bearing architectures capable of withstanding automated adversarial attacks.