The agentic frontier: A risk and resilience framework for autonomous AI reasoning
The Internet is shifting from a human-centric web towards an agentic frontier, with artificial intelligence (AI) agents increasingly integrated into daily life. The transition from instruction-based bots to goal-orientated agents is accelerating, signalling a move from automation to autonomy. According to internal Akamai data, AI bot traffic grew by 300 per cent over the past year, bringing a new kind of Internet traffic characterised by distinct behaviours. As AI agents evolve, existing security defences are proving inadequate; signature-based rules cannot effectively counter reasoning-driven agents. This paper examines the taxonomy of attacks, both AI as a threat and threats targeting AI systems. Whether dealing with aggressive AI inference crawlers such as Bytespider or innovative AI-engineered threats, current security limitations heighten business risks and operational challenges. This necessitates a fundamental evolution, not only of traditional firewalls but also of the security architecture, towards a new, agent-aware defence framework that moves beyond identifying who is accessing resources to understanding the intent and logic of autonomous actors. Semantic firewalls are essential, protecting users and agents at input and output stages while keeping models safe. To enable verifiable autonomy, strategic shifts at the board level are critical, establishing governance frameworks that allow organisations to transition from black-box models to transparent, verifiable glass-box models, paving the way for trustworthy autonomous systems. This article is also included in The Business & Management Collection which can be accessed at https://hstalks.com/business/.
Authors
- Ryan Gao (ORCID: https://orcid.org/0000-0002-7030-1170)
- Karan Mankodi
Institutions
- Akamai (United States) (US)
Publication Details
- Journal
- Cyber security.
- Published
- 2026-10-06
- DOI
- https://doi.org/10.69554/dgay5847
- Primary Topic
- Adversarial Robustness in Machine Learning
- Type
- article
- Field-Weighted Citation Impact
- 0.00