Claude is gaining a new window into every agent turn, model call, and tool use. On October 14, 2026, Anthropic announced OpenTelemetry trace export for the Claude Platform, giving developers a standard way to send Claude activity into observability tools while the company expands a cyber mission aimed at finding and fixing vulnerabilities.
The two moves point toward the same challenge: AI systems are becoming more capable, but the infrastructure around them must become easier to inspect and protect. Developers can now connect Claude Managed Agents, the Messages API, and Claude SDKs to observability tools that accept the OpenTelemetry standard.
Claude Activity Becomes Easier to Follow
Adding a destination in the Claude Console sends every agent turn, model call, and tool use to the selected observability tool as a standard OpenTelemetry trace. Claude traces follow the OpenTelemetry GenAI semantic conventions, creating a shared structure for understanding how an AI request moves through a system.
Anthropic defines a trace as the full record of one request through a system, while a span represents one step within that trace. That distinction gives teams a way to examine the complete request and then inspect the individual actions that produced it.
- Managed Agents and the Messages API require a destination configured in the Claude Console.
- Claude SDKs export traces automatically after developers install the SDK and register a tracer provider.
- SDK tracing is configured in code and requires no special permission.
- API tracing requires adding a destination in the Claude Console, which requires Admin permission.
- Each span can hold up to 0.75 MB of content by default.
For teams building with Claude, this creates a clearer path from an AI request to the model calls and tools used along the way. The data can flow into observability systems that already accept OpenTelemetry, so developers can examine agent behavior through a standard format instead of treating every platform as a separate monitoring problem.
Anthropic’s Cyber Mission Puts AI Against Vulnerabilities
Anthropic’s cyber initiative adds a much bigger layer to the story. The company announced the Cyber Mission on October 8, 2026, with Anthropic models and engineers moving through three tracks: the Critical Infrastructure Defense Program, the OSS Scanner, and the expanded Cyber Verification Program.
Anthropic’s disclosure dashboard lists 29,439 findings discovered between November 1, 2025, and October 2, 2026. As of October 2, 2026, 6,157 findings were disclosed across 591 open-source projects, with all 6,157 findings reported to maintainers and 516 patched upstream to Anthropic’s knowledge.
The company also says its findings were reviewed by six security firms. The Cyber Mission reports 6,123 findings reviewed, with 5,674 confirmed valid, producing a 92.7% true-positive rate. Of the 6,157 reported findings, 1,333 came through triage and 4,824 went directly to maintainers.
The pace has changed fast. Findings reported to maintainers rose from 1,596 to 6,157, a 3.9x increase, while findings patched upstream grew from 97 to 516, a 5.3x increase. Patched findings also rose from 6.1% of reported findings to 8.4%, though Anthropic said a patch does not guarantee widespread installation.
That final gap matters. Finding a vulnerability is one task; getting every affected operator to install the fix is another. Adam Meyers described the pressure this way: “the problem is patching. At ten times today’s findings, they’re going to be completely underwater.”
Critical Infrastructure Faces the Hardest Test
The Critical Infrastructure Defense Program has 11 founding partners: Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC, and Rockwell Automation. The program arrives as water and wastewater utilities in at least seven states reported attacks on controllers since July 27, 2026, with consequences including loss of pressure and flooding.
On July 30, 2026, the FBI and EPA warned operators to remove programmable logic controllers from direct internet exposure by using secure gateways and firewalls. Tom Dobbins of WaterISAC said, “OT systems that are exposed to the internet are a major challenge, and many of these systems that are older generation need to be not accessible to the internet.”
The OSS Scanner provides free periodic scans to enrolled projects, with reports sent without human review. Anthropic did not specify whether partners will receive free model access or who will cover computing costs, leaving an important question for organizations that need sustained scanning and remediation.
Anthropic’s June program for government entities reached more than half of all U.S. states, but access alone does not solve the operational problem. Cynthia Kaiser said, “it’s not just about getting access, and that the question is where do they start?”
That concern reaches local communities with limited budgets. A $100 million bill would authorize operators’ frontier-model access if funded by appropriators, while Josh Gottheimer’s AI Cyber Defense Act would authorize $100 million for a CISA pilot from 2027 to 2031. Gottheimer put the resource gap plainly: “Many of our local communities just don’t have the resources they need to pay for AI tokens to do the patching they need.”
Claude’s new trace export and Anthropic’s cyber programs now form two sides of a larger push: make AI activity visible, then use AI to help expose weaknesses across software and infrastructure. The next test will be turning that visibility into patches that reach the systems that need them most.
Based on
