Claude Opus 5 Tops Artificial Analysis Index While Maintaining Strict Cybersecurity Safeguards
Anthropic has launched Claude Opus 5, a new flagship model that tops the independent Artificial Analysis Intelligence Index while incorporating deliberate restrictions on offensive cybersecurity tasks.
Released on 24 July, Opus 5 is priced at half the cost of the restricted Fable 5 and is available to all users without export-control limitations that previously affected earlier models. The model supports a 1,000,000-token context window by default and generates up to 128,000 output tokens, with reasoning enabled by default through five effort levels: low, medium, high, extra, and max.
On the Artificial Analysis Intelligence Index, which aggregates nine benchmarks including Terminal-Bench, SciCode, GPQA Diamond, and Humanity’s Last Exam, Opus 5 scored 61 points at maximum effort, placing first among 187–191 models. It outperformed Fable 5 (60) and GPT-5.6 Sol (59). However, the model consumed approximately 100 million tokens during evaluation compared with the median of 63 million, resulting in slower response times exceeding one minute to first token.
Internal Anthropic benchmarks showed strong gains: 43.3% on Frontier-Bench v0.1 (versus 33.7% for Fable 5), 1861 Elo on GDPval-AA v2, and a notable 30.2% on ARC-AGI-3. The model also led on OSWorld 2.0 with 70.6%. It underperformed on DeepSWE v1.1 (68.8% vs 72.7% for GPT-5.6 Sol), HealthBench Professional, and Legal Agent Benchmark.
In cybersecurity testing, Opus 5 reached 79.4% on OSS-Fuzz vulnerability discovery, nearly matching Mythos 5 at 80%. Exploit generation success remained low at only four tasks versus 13 for Mythos 5. The model was not trained on offensive cybersecurity and routes blocked requests for binary scanning, penetration testing, or exploit generation to Opus 4.8.
Adjustable effort levels allow users to balance performance and cost, with Anthropic recommending extra for coding and agentic work and high for general tasks. The model’s tendency toward extensive self-verification improves reliability on complex tasks but can cause unproductive overthinking on certain autonomous research scenarios.
Related articles
OpenAI GPT-6 Astra Deploys Multi-Agent Parallel Processing, Increasing Local CPU Load and Security Risks
Early users of GPT-6 Astra have observed the model distributing complex tasks across multiple specialized agents that plan, solve, test code, verify results, and iterate after failures. This multi-agent approach enables faster handling of multi-step workflows compared to sequential chatbots. OpenAI states that Astra can control computers, operate browsers and applications, and install or test software, though it has not officially confirmed a native multi-agent architecture. Main computations run in the cloud, but agent tools can execute on user devices or corporate servers, leading to noticeable processor load when multiple agents compile code, launch browsers, run tests, and operate containers simultaneously. Corporate environments face added complexity as each agent requires virtual machines, sandboxes, internal data access, and careful environment cleanup. The increased autonomy has prompted OpenAI to strengthen monitoring of Astra actions and permission boundaries for subscribers of ChatGPT and enterprise clients.
Microsoft Copilot Can Surface Overshared Data Despite Permission Boundaries
Microsoft documentation states that Copilot only accesses data authorized for the signed-in user, yet default SharePoint and OneDrive sharing settings often grant broad access that the AI then respects literally. This creates accidental oversharing risks where Copilot retrieves documents shared too widely years earlier. Administrators can use Content Management Assessment and Data access governance reports, including the EEEU report covering the top 100 sites shared in the past 28 days, to identify problematic content. Two distinct controls exist: Restricted Access Control removes access entirely while Restricted Content Discovery hides items from Copilot and search without altering permissions. Sensitivity labels combined with encryption can exclude programmatic access for agents, though Microsoft does not guarantee outright blocking. Interaction logs stored in Microsoft Purview retain user prompts, Copilot responses, and citations to accessed documents, providing an audit trail for oversharing incidents.
Adaptive LLM Worm Uses Local Models to Craft Per-Target Exploits in Heterogeneous Networks
Researchers from the University of Toronto have published a preprint describing an adaptive computer worm driven by LLM agents that spreads across corporate networks by generating individualized attack strategies for each compromised system. Unlike traditional worms such as WannaCry that rely on fixed exploits, this worm maintains its own infrastructure by running local LLMs on infected GPU-equipped machines to analyze vulnerabilities and synthesize new attack vectors in real time. The system was tested in an isolated FakeCorp environment containing Linux, Windows, and IoT devices, successfully leveraging known real-world vulnerabilities to propagate over 48 hours and seven-day autonomous runs. Two core components power the worm: a GPU-hosted LLM component and a hierarchical agent framework with memory, reasoning graph, and tool modules that manage reconnaissance, exploitation, and payload deployment. The authors note that the approach creates an economic asymmetry favoring attackers because the worm parasitizes victim compute resources, eliminating the need for external C2 or commercial LLM services. They warn that adding adaptive reasoning to historical worms such as SQL Slammer, Conficker, or Stuxnet would significantly increase their resilience while remaining slower and noisier than classic self-propagating malware.
Building Secure On-Prem AI Assistants: How to Keep Corporate Data Inside Closed Contours
Many organizations hesitate to deploy AI assistants due to strict data protection rules that prohibit sending information to external clouds. The article explains how to implement AI models entirely within a company's own infrastructure, ranging from on-premise servers to fully offline laptops. It breaks down four deployment locations from public APIs to local devices and clarifies three distinct access levels: read, write, and execute. The author emphasizes that most business value comes from read-only access combined with human-in-the-loop controls for any irreversible actions. Practical recommendations include RAG over model size, quantization for local hardware, and maintaining immutable audit logs. The piece also warns that preparing clean knowledge bases often consumes more effort than the model itself.