Anthropic’s new artificial intelligence is blocked from public access due to security risks

Developer Anthropic announced the public blocking of its new artificial intelligence model, Claude Mythos Preview, this Tuesday, April 7, 2026. The tool presented advanced coding and logical inference capabilities during initial evaluations, surprising the global technology market. The company’s management chose to restrict the use of the system to only a select group of strategic partners. The central reason for the decision involves the high risk of exploiting vulnerabilities in the internet infrastructure.

The software recorded 93.9% accuracy on the SWE-bench Verified indicator. Esse index represents a world record in automated software engineering efficiency. Durante stress tests, the technology identified thousands of zero-day failures, which are loopholes unknown to the original code creators themselves. Unrestricted release of the program would allow the creation of destructive cyber attacks on a global scale. The governance of the model will be the responsibility of a closed corporate council to prevent technology leakage.

Descoberta of historical vulnerabilities in open source systems

Claude Mythos Preview’s analytical capabilities have surpassed traditional methods of manually reviewing complex programming. Artificial intelligence located a critical flaw in the OpenBSD system in the first weeks of laboratory operation. The security error had remained hidden and uncorrected for almost three decades. Especialistas humans had not detected the issue in previous routine audits. Massive data processing has proven to be essential for mapping these deep layers of vulnerability on platforms considered robust.

Fault mapping also reached FFmpeg, a worldwide video processing software. The model found a structural bug that had existed on the platform for 16 years. The tool demonstrated technical accuracy in classifying the remote code execution vulnerability in the FreeBSD system. The defect was officially classified as CVE-2026-4747 in international security registries. The speed of detection indicates a change in the industrial standard, where error localization occurs almost instantly.

Formação of Projeto Glasswing and Enterprise Access Control

The administration of the new artificial intelligence model occurs through Projeto Glasswing. The initiative functions as a digital defense coalition focused on protecting critical infrastructure against future automated threats. The restricted group is allowed to apply Claude Mythos Preview processing to search for defects in their own networks and hardware. The main strategy aims to patch the holes before malicious actors independently develop similar attack mechanisms.

The technical protection network involves more than 40 organizations from different sectors of the global economy. The cooperation integrates direct competitors in the technology market and large financial institutions. The technical council has established strict guidelines for using the analytical tool:

• Amazon Web Services and Microsoft Azure use the system to audit cloud servers and seal digital perimeters.
• Empresas like Google, Apple and NVIDIA apply the technology to scan their own hardware and operating systems.
• JPMorgan Chase leads the implementation of fixes in banking networks and financial transaction systems.
• Cisco, Linux Foundation, and Broadcom oversee the transformation of alerts into real-time security updates.

Cybersecurity companies CrowdStrike and Palo Alto Networks work to convert the data generated by the system. Institutions translate artificial intelligence discoveries into direct protection signatures for end customers. The process creates a digital shield that benefits millions of users indirectly. Consumers receive security fixes without having to interact with the original algorithm. Access to the model requires compliance with protocols that prevent the export of raw data, resembling treaties controlling technologies for military use.

Leia Tambem

Nintendo reveals new hybrid console with Nvidia processor and remake of Zelda Ocarina of Time

Sony confirms removal of Red Dead Redemption and five games from PS Plus Extra and Deluxe in June

New release date for Grand Theft Auto VI is set for November 2026 by the producer

Comparativo process performance and handling technician

Laboratory evaluations in modern browsers have highlighted the abysmal technical difference between generations of artificial intelligence. Engineers tested the systems in simulated Firefox 147 environments. The previous model, Claude Opus 4.6, managed to perform just two shell exploits out of hundreds of attempts. Claude Mythos Preview was successful in 181 attacks under the same technical stress conditions. The evolutionary leap requires rapid adaptations from cyber defense teams to contain new attack realities.

Control of system records showed results that required immediate action from Anthropic developers. The new version took command of critical processes 29 times during security simulations. The metric demonstrates that the algorithm understands manipulating the operational core of systems, not just locating the gateway. The company established containment protocols in isolated environments, technically known as air-gapped. The measure prevents artificial intelligence from communicating with external servers without express authorization from supervisors.

Impacto in legacy software maintenance and market projections

The system architecture allows simultaneous analysis of millions of lines of code for memory leaks and stack overflows. The tool understands the execution context and drastically reduces the incidence of false positives during scans. The effectiveness in JavaScript environments indicates that the current structure of the web requires urgent reformulation. The model can easily bypass sandboxing protections in browsers. Digital isolation barriers will undergo complete overhauls to support the new level of automated inspection.

The application of technology directly benefits the maintenance of legacy software. Muitos industrial control systems and power grids operate with code developed in the 1990s and 2000s. Administrators avoid frequent updates due to the risk of instability in daily operations. Claude Mythos Preview audits these legacy infrastructures in a secure and controlled manner. The proposed modifications increase protection without interrupting essential services provided to the population.

The Anthropic board issued a warning indicating that the model’s technical exclusivity has a short expiration date. The accelerated advancement of hardware will facilitate the development of analogous systems by other entities soon. The technology industry uses the current period to promote a transition to inherently resilient software architectures. Human programming will receive constant validation from security algorithms in the near future. The strategy focuses on ensuring that patch packages reach users before exploit tools become accessible outside of the labs.