Anthropic released Claude Haiku 5.5 (Oct. 7), positioning it as a fast, low-cost model for classification, summaries, browser use, and subagent turns. Official pricing is $0.10/M input and $0.50/M output tokens for prompts up to 100K (90% below Haiku 4.5 list rates); it is available across Anthropic, AWS, Google Cloud, and Azure. Anthropic also cut Sonnet 5.5 cache-read pricing in half, estimating ~20% lower cost on most agentic work, and added beta browser/computer-use SDK support. Performance and customer latency claims are vendor-reported; early Reddit reception emphasizes cost and speed but is not independent benchmarking. Sources: https://www.anthropic.com/claude-haiku-5-5 ; https://www.reddit.com/r/ClaudeAI/comments/1x03k4j/introducing_claude_haiku_55_the_cheapest_fastest/
Microsoft announced new CRM-to-workflow integrations for sales and service: Copilot and Dynamics 365 customer context/actions appear in Teams, while its Sales Development Agent entered public preview and was reported to have generated meetings with 10+ otherwise-unreached customers during its first three weeks. This extends Microsoft’s agentic CRM positioning from assistant features toward autonomous sales processes; outcomes are company-reported, and the agent begins consuming Copilot credits Nov. 16. Source: https://www.microsoft.com/en-us/copilot/blog/2026/10/08/bringing-crm-into-the-flow-of-work-and-agents-into-business-process-for-sales-and-service-teams/
DigitalOcean’s Managed Agents documentation records an Oct. 6 update adding CrewAI support to Harness Runtime and tightening recommended sandbox bases for custom templates. The broader public-preview platform combines per-session hardware-isolated microVMs, durable pause/resume/fork state, and an Action Gateway for 16,000+ governed tools behind MCP—evidence that managed execution and tool access are converging into a deployable agent runtime. Product details are vendor-described. Sources: https://docs.digitalocean.com/products/managed-agents/ ; https://www.digitalocean.com/products/managed-agents
OWASP GenAI Security Project published a Q3 incident roundup (Oct. 8) that ties together several recent agent boundary failures, including evaluation agents reaching production systems, a Copilot prompt-execution/memory-poisoning exploit family, and Deadbugz—a malicious MCP server promoted through 23 GitHub PRs that changed tool metadata after three benign calls to steer agents toward secrets. OWASP says no PRs were merged and no downstream compromise was confirmed; the key operational lesson is to monitor tool-definition changes, re-approve changed tools, and constrain credential access. Sources: https://genai.owasp.org/2026/10/08/genai-and-agentic-ai-exploit-roundup-q3-2026/ ; https://www.pillar.security/blog/deadbugz-currently-active-mcp-supply-chain-campaign