Anthropic researchers warn agentic AI cyberattack risks could reach 10%

The shift from passive chatbots to autonomous agents represents a significant expansion of the threat landscape for digital infrastructure. By connecting large language models to active tools like web browsers and code executors, the technical barrier for sophisticated hacking is lowered, allowing non-experts to launch automated swarms. For the streaming and tech ecosystem, this necessitates a move away from 'kill switch' legislation toward mandatory auditing of internet-capable agents before deployment. As companies like Anthropic and OpenAI race to market, the industry must prioritize transparent incident reporting over marketing-driven hype. Watch for whether the U.S. government establishes new export controls on high-tier cybersecurity models similar to weapons technology.

Anthropic's warnings about agentic AI cyberattack risks arrive as telecom operators and cloud providers accelerate production deployments of autonomous agents across network infrastructure. In June 2026, Ericsson launched its AI in RAN commercial software subscription claiming up to 20% higher downlink throughput across more than 15 live deployments, while Verizon disclosed that its 60,000-site vRAN network is now applying agentic AI to configuration changes, service assurance, and optimization. These deployments demonstrate that autonomous agents are already operating at scale in critical infrastructure, raising the stakes of the safety failures Anthropic researchers describe.

Nokia has moved aggressively to position itself at the center of the agentic AI value chain for telecom. The company announced partnerships with AWS and Databricks to build a unified data and control layer for autonomous networks, with its Autonomous Network Fabric running on AWS from later in 2026. Nokia claims operators using its autonomous networks portfolio are achieving automation rates above 90 percent and service delivery times of four hours or less. However, Nokia executives acknowledged the risks of deploying AI agents without sufficient guardrails, keeping humans in the loop until a zero-trust environment is established. The company's mobile core team reported that agentic AI reduced call setup times from roughly 10 seconds to one or two seconds in certain use cases, but maintained human supervision as a prerequisite for production rollout.

The competitive divergence between Nokia and Ericsson on AI architecture has direct implications for how safety oversight is structured. Nokia's RAN strategy is now built on its partnership with Nvidia, with the chipmaker's $1 billion investment anchoring a GPU-based approach, while Ericsson pursues incremental intelligence on existing baseband silicon. T-Mobile US is testing both approaches simultaneously, highlighting the split between incremental RAN intelligence and broader shared AI compute models. This architectural fragmentation means that any regulatory framework for agentic AI safety must account for fundamentally different deployment topologies, from GPU-accelerated edge inference to centralized cloud-hosted agents, each presenting distinct attack surfaces for the kinds of automated exploits Anthropic researchers flagged.