01
Claude Opus 5 launches with stable token pricing
Anthropic released Claude Opus 5 as its newest flagship model and kept pricing at $5 per million input tokens and $25 per million output tokens. Anthropic made Opus 5 available in Claude products and via the API.
- Fixed token pricing simplifies vendor comparisons and budget modeling versus OpenAI, Google, and Microsoft-managed stacks.
- A single flagship available in both UI and API reduces divergence risk between end-user deployments and embedded applications.
- Procurement teams can align rate-card governance early because model upgrades do not automatically change unit pricing.
02
Anthropic discloses three investigated cyber incidents
Anthropic published details of three real-world cybersecurity incidents it investigated where AI systems were used in cyber activity, including cases involving its models. The company positioned the write-up as part of its cybersecurity evaluations work.
- CISOs can use the incident narratives to stress-test internal controls for agent workflows, including approval gates and least-privilege tool access.
- The disclosure provides concrete inputs for risk assessments, audit logging requirements, and red-team testing plans before expanding AI automation.
- Security teams can validate whether their vendor due-diligence questionnaires cover incident response, investigation process, and transparency commitments.
03
MCP 2026-07-28 rolls out with stronger authorization
Anthropic announced MCP 2026-07-28 support across Claude products, moving MCP to a stateless core and hardening authorization while graduating official extensions. The company described the update as part of ongoing product rollout.
- A stateless core can simplify enterprise architecture reviews by reducing state-handling assumptions inside connectors and integrations.
- Harder authorization directly affects how Czech enterprises implement access controls for internal systems connected to Claude (HR, ERP, ticketing).
- Graduated official extensions can reduce reliance on ad-hoc connectors and make integration governance easier to standardize across business units.
04
Sonnet 5 positions as Anthropic’s agentic work model
Anthropic shipped Claude Sonnet 5 and described it as its most agentic Sonnet model yet, emphasizing planning, tool use, and autonomous work. Anthropic made the model available across Claude plans and via the platform.
- IT leaders can evaluate whether a lower-tier model meets internal coding and knowledge-work needs before committing to flagship spend.
- Agentic tooling claims raise governance requirements for tool permissions, action logging, and change control in production workflows.
- Broad plan availability matters for standardized rollout patterns (dev teams, analysts, and support) without fragmenting model policy.
05
Claude Code adds Opus 5 as default Opus model
A Claude Code release update added Claude Opus 5 support and set it as the default Opus model in the tool, alongside SDK-related updates. The change affects which model teams use when they select an Opus option in Claude Code.
- Default model changes can alter cost, latency, and output behavior, so engineering leads should pin versions for reproducible builds and evaluations.
- Security and compliance teams should confirm logging and data-handling settings in developer tools because model upgrades often drive renewed usage.
- Platform owners can use the update as a trigger to refresh internal guidance for allowed models, rate limits, and code review policies.
06
NYT spotlights Anthropic cyber incident disclosure and safety posture
The New York Times reported on Anthropic’s incident write-up and discussed broader scrutiny of model behavior and safety positioning. The coverage amplified the enterprise risk-management angle around AI-enabled misuse.
- Public scrutiny increases the likelihood of deeper vendor risk reviews, which can extend procurement timelines for regulated Czech buyers.
- CIOs can use the discussion to align stakeholders on required safeguards for agentic deployments, especially around tool access and monitoring.
- Communications and legal teams should prepare incident-response messaging for AI systems, because external narratives can shape internal escalation thresholds.