Machine traffic has officially overtaken human activity on the internet. According to RisingInfo's 2026 Automated Threat Report, bots now account for 68% of all web traffic, while genuine human visits have dropped to just 22%. More striking: AI Agent-driven requests have exploded from under 1% in early 2025 to 8-12% by Q2 2026, with total LLM and Agent requests exceeding 450 billion—a 400% year-over-year surge.
This isn't merely a shift in volume. It represents a fundamental transformation in how the internet operates. For decades, security models assumed a binary world: humans (legitimate) versus bots (suspicious). That framework is collapsing. We now face what the report calls a "tri-body traffic" ecosystem—three distinct actors coexisting, each requiring different treatment.
Three Species, One Network
The report categorizes non-human traffic into three evolutionary stages:
- Traditional Bots: Rule-based automation—scrapers, credential stuffers, inventory hoarders. Predictable but persistent.
- AI-Enhanced Bots: Machine learning-augmented tools that adapt their tactics based on defensive responses.
- AI Agents: Autonomous systems capable of understanding objectives, planning multi-step paths, calling APIs, and adjusting behavior based on environmental feedback. These don't follow scripts—they pursue goals.
This taxonomy matters because each type demands a different security response. Blocking all automation is no longer viable when legitimate AI Agents handle business-critical workflows. The question shifts from "Is this a machine?" to "Should I trust this machine, and for what purpose?"
The L1-L5 Threat Spectrum
RisingInfo proposes a five-level framework mapping threat sophistication:
L1-L2 (Tool-Based Attacks): Low autonomy, predefined rules. Traditional DDoS, basic scraping.
L3 (Adaptive Automation): ML-enhanced evasion, behavior mimicry.
L4 (Goal-Driven Agents): Autonomous planning with specific objectives. Can navigate complex authentication flows, solve CAPTCHAs via vision models, and persist across sessions.
L5 (Coordinated Multi-Agent Systems): Swarms of agents collaborating, sharing intelligence, and orchestrating attacks across multiple vectors simultaneously.
The report identifies four new AI-specific threat vectors: LLM application attacks, Agent supply chain compromises, Agent identity hijacking, and AI-autonomous attack orchestration. Financial services now see 60% malicious bot traffic. Manufacturing is experiencing the fastest growth in automated threats.
From "Block Bots" to "Govern Trust"
The security paradigm is inverting. Traditional WAAP (Web Application and API Protection) focused on detection and blocking—identifying signatures, blacklisting IPs, challenging suspicious requests. This approach fails when:
- AI Agents use legitimate credentials and browser environments
- Traffic originates from residential proxies and cloud infrastructure indistinguishable from business use
- Behavioral patterns mirror human interaction (mouse movements, scroll depth, timing)
The proposed alternative: trust governance. Rather than binary allow/block decisions, systems should evaluate identity, behavior patterns, and intent to assign dynamic trust scores. High-trust agents receive seamless access. Medium-risk flows trigger step-up authentication. Low-trust traffic faces strict verification or rejection.
This mirrors broader shifts in identity management—from perimeter-based security to zero-trust architectures. The machine layer requires similar evolution.
Implications for Infrastructure Builders
For platform architects and security engineers, this transition creates several immediate requirements:
Identity for Non-Humans: Agents need verifiable credentials—signed attestations of their origin, capabilities, and authorized scope. The concept of "machine identity" must mature beyond API keys to cryptographic proofs and attestation frameworks.
Behavioral Baselines: Static rules fail against adaptive agents. Security systems need behavioral ML models that establish normal patterns for different traffic types and flag deviations without blocking legitimate automation.
Agent-Friendly Channels: Organizations should create designated pathways for verified AI Agents—structured APIs, documented rate limits, and clear terms of service. Making legitimate agent access easier reduces incentives to evade detection.
Continuous Assessment: Trust scores must decay over time and adjust based on ongoing behavior. An agent that was legitimate yesterday may be compromised today.
What You Can Do Now
For operators running production systems:
Audit your current bot management: Does your WAF distinguish between beneficial automation (search crawlers, monitoring systems, business logic agents) and threats? Or does it apply blanket rate limits that damage legitimate operations?
Implement graduated responses: Design traffic handling that varies by trust tier—fast path for verified agents, standard flow for ambiguous traffic, challenge-interstitial for suspicious patterns, and block for confirmed threats.
Prepare for agent identity standards: Monitor emerging specifications like Agent Identity Protocol and Agent Trust Framework. Early adoption of interoperable standards will provide competitive advantage as agent ecosystems mature.
Review data exposure: AI Agents excel at extracting information through multiple small requests that individually appear benign. Assess whether your APIs, when queried systematically, expose sensitive patterns or enable reconstruction of proprietary data.
The New Internet Topology
The "tri-body" framing extends beyond security. It describes the emerging architecture of digital interaction: humans browsing, bots performing routine automation, and AI Agents pursuing complex objectives across systems.
This configuration isn't temporary. As agent coordination frameworks mature and compute infrastructure scales, machine-initiated activity will grow as a share of total traffic. The organizations that thrive will be those that design for this reality—not by treating all automation as threat, but by building infrastructure that distinguishes, authenticates, and appropriately privileges different actor types.
The internet began as a human-to-human network. It evolved into a human-to-machine system. It's now becoming a machine-to-machine ecosystem with humans as one participant among three. Security architecture must evolve accordingly.
