Kite AI Details Security Vulnerabilities in 'Agentic Internet'
Image Source: Go Kite
Kite AI, a startup building infrastructure for autonomous AI agents, has highlighted security risks in the developing "agentic internet", where AI systems equipped with memory and identity could be vulnerable to attacks, potentially endangering data privacy in API-based collaborations.
In a thread on social media platform X dated July 16, 2025, the company outlined threats to these agents, which are engineered for independent operations.
Background on the Agentic Internet
The agentic internet envisions a move away from human-led digital interactions toward ecosystems led by autonomous AI agents. These leverage progress in large language models and lower compute expenses to authenticate identities, analyze data and perform actions absent continuous human input.
Kite AI's insights arise during industry momentum for agentic AI. IDC anticipates such technologies will comprise over 26% of worldwide IT expenditure by 2029. The firm specializes in core components like cryptographic identities and governance structures to enable this change.
Established by Chi Zhang and Scott Shi, Kite AI merges AI with blockchain. Zhang, a University of California, Berkeley AI PhD holder, formerly directed data products at Databricks; Shi engineered real-time AI systems at Uber. The company runs a Layer 1 blockchain on Avalanche with a Proof of Artificial Intelligence consensus, facilitating agent transactions in stablecoins at low costs.
Security Challenges Identified
Kite AI specified four threats in its thread. Memory tampering entails injecting inaccurate data into an agent's retained context, risking flawed outcomes like incorrect financial dealings. Identity spoofing permits imposters to mimic authorized agents, weakening verification.
Data poisoning happens via corrupted API inputs to agents, while adversarial inputs capitalize on model flaws to provoke errant behaviors. Such issues derive from agents' enduring memory and self-reliance, positioning them as exploitable entities in linked setups.
The analysis underscores escalating privacy concerns as AI agents interconnect across services without human checks.
Proposed Solutions and Technical Approaches
Kite AI suggested agent-tailored safeguards extending past standard web protections. For memory tampering, it proposed immutable logs via SHA-256 hashing and Merkle trees for data assurance.
Identity spoofing countermeasures include zero-knowledge proofs alongside agent registries for authentication. For data poisoning, federated validation and trust scoring were advised; adversarial inputs call for bolstered model resilience and live anomaly spotting.
Further tactics encompass end-to-end encryption in exchanges, provenance for tracing data roots, behavior monitoring for irregularities and sandboxing to quarantine issues. Kite AI advocated incorporating these inherently, fusing cryptography and blockchain for solid trust frameworks.
Company Developments and Funding
Kite AI's observations complement its portfolio, featuring agent discovery tools, permission oversight and stablecoin transactions. It maintains a partnership with PayPal in pilot phase and an API integration with Shopify, enabling merchant access through its agent app store.
On Sept. 2, 2025, Kite AI raised US$18 million in Series A funding spearheaded by PayPal Ventures and General Catalyst, elevating total funds to US$33 million. Backers encompass Samsung Next, HashKey Capital, Animoca Brands and angels from OpenAI, AWS and academia. The investment aids network growth, with transaction fees under $0.000001 and one-second block intervals.
Latest alliances include one with Vishwa for agentic liquidity enhancements. The platform logs metrics such as 1.01 million highest daily agent interactions, 1.7 billion agent interactions and 17.8 million agent passports.
Impacts and Future Trends
If ignored, these vulnerabilities might impede uptake, worsening problems in extant AI uses like breaches in generative platforms. In security realms, agentic AI could heighten risks such as automated phishing yet foster superior safeguards, per specialists at Infosecurity Europe.
Projections indicate the AI agents market reaching US$243.7 billion by 2035, spanning e-commerce and research applications. Efforts like Salesforce's Agentforce and Databricks' initiatives highlight surging commitments.
Emerging patterns favor AI-blockchain hybrids for trust bolstering, possibly redefining finance and healthcare. Still, oversight on data privacy via GDPR and CCPA will guide evolution, affecting compliance and control norms. Kite AI's stance, from an external lens, underscores secure decentralized foundations as key to the agentic internet's sustainability, harmonizing advancement with hazard control.
TheDayAfterAI News
We are a leading AI-focused digital news platform, combining AI-generated reporting with human editorial oversight. By aggregating and synthesizing the latest developments in AI — spanning innovation, technology, ethics, policy and business — we deliver timely, accurate and thought-provoking content.