Anthropic Launches Claude Sonnet 5 and Restores Access to Fable and Mythos

Thailand AI News: Anthropic has unveiled Claude Sonnet 5 while restoring full access to its Fable and Mythos frontier AI models following an intensive federal export control review that temporarily halted availability of some of its most advanced artificial intelligence systems. The announcement marks an important moment for both the company and the wider AI industry, highlighting how rapid technological innovation is increasingly being shaped by security concerns, government oversight and growing enterprise demand for reliable AI platforms.

The restoration comes after an eighteen-day operational suspension that began on June 12 when US export control measures required Anthropic to temporarily disable access to Fable 5 and Mythos 5 while engineers addressed a reported security vulnerability. Midway through this Thailand AI News report, it becomes clear that the incident was about far more than a single software flaw. It demonstrated how frontier AI models are entering an era where regulatory scrutiny is becoming as important as technical performance. Following extensive testing and the introduction of new safeguards, Anthropic has now resumed commercial access across its platform, cloud services and authorized partner ecosystems.

Security Review Prompted Temporary Shutdown

The suspension followed research by Amazon that identified a method capable of bypassing safety controls within Fable 5. During testing, the exploit allowed the model to identify software vulnerabilities and generate code that could potentially be used to exploit them, prompting federal authorities to intervene while the issue was investigated.

Without a real-time nationality verification system that could satisfy the new export requirements, Anthropic chose to suspend access globally rather than risk breaching the directive. Although the interruption affected developers and enterprise customers worldwide, it also gave the company an opportunity to strengthen its safety architecture before returning the models to production.

Engineers responded by creating a new automated safety classifier specifically designed to detect the prompt pattern identified during Amazon’s research.

Anthropic said internal validation showed the updated protection blocked the reported exploitation method in more than 99 percent of testing scenarios.

When the classifier identifies a request that falls within the new safety boundary, the platform automatically redirects the task to the older Claude Opus 4.8 model. While the wider safety margin may occasionally interrupt legitimate software debugging sessions, Anthropic believes the additional protection is necessary as increasingly capable AI systems become more widely deployed.

Claude Sonnet 5 Targets Enterprise Adoption

While attention has centered on the restored frontier models, Anthropic’s commercial focus is firmly on Claude Sonnet 5, which has been designed to deliver strong performance while reducing operating costs for businesses deploying autonomous AI agents.

Benchmark results indicate noticeable gains over Claude Sonnet 4.6. The model achieved 63.2 percent on SWE-bench Pro, compared with 58.1 percent for its predecessor, while recording 80.4 percent on Terminal-Bench 2.1 against the earlier model’s 67 percent. Although Claude Opus 4.8 remains Anthropic’s highest-performing model, Sonnet 5 offers organizations a more cost-effective balance between performance and affordability.

Standard pricing remains at US$3 per million input tokens and US$15 per million output tokens, with introductory pricing reducing those costs to US$2 and US$10 respectively until August 31, 2026.

Several major technology companies have already deployed the model. Rakuten is using Claude Sonnet 5 to review production code, execute tests and validate software before presenting completed work to engineers. Zapier has integrated the system into enterprise workflows that involve updating Salesforce accounts before automatically preparing and distributing customer announcements, while development platform Zed has successfully used the model to diagnose software bugs, generate testing scripts and apply verified code fixes with minimal human involvement.

Safety Measures Expand Alongside AI Capabilities

Anthropic has emphasized that stronger performance has not come at the expense of safety. According to the company’s latest system card, Claude Sonnet 5 demonstrated lower levels of non-compliant behavior than Claude Sonnet 4.6 during internal behavioral testing, despite offering improved reasoning and coding capabilities.

The company deliberately excluded specialized offensive cybersecurity datasets from the model’s training process, limiting its focus to defensive software engineering and general programming tasks. Additional assessments carried out in partnership with Mozilla examined whether the model could produce working exploits for known vulnerabilities within the Firefox 147 browser. Anthropic reported that Claude Sonnet 5 failed to generate a functional exploit during testing, recording a zero percent success rate. Although researchers observed a modest increase in partially successful responses compared with earlier models, engineers attributed the difference to stronger logical reasoning rather than any deliberate enhancement of offensive cyber capabilities.

Commercial deployments also benefit from the same real-time safety classifier technology used with Anthropic’s flagship Claude Opus 4.8 model, providing continuous monitoring for potentially harmful requests while allowing legitimate development work to continue.

Industry Collaboration on AI Security

The temporary suspension of Fable 5 has encouraged closer cooperation across the AI industry as companies work towards common standards for evaluating emerging security risks.

Anthropic has joined Amazon, Microsoft and Google in developing a shared framework that will help determine the severity of AI model vulnerabilities. The proposed system evaluates four key factors: the level of capability gained through an exploit, the range of harmful activities it enables, the ease with which it can be weaponized and how widely the technique is known within the research community.

The objective is to establish consistent criteria for responding to newly discovered vulnerabilities rather than leaving each developer to determine its own assessment. High-risk incidents affecting financial systems, critical infrastructure or other essential services would trigger immediate defensive measures, helping reduce the potential impact before corrective updates are released.

Anthropic has also strengthened its broader security programme through a HackerOne vulnerability disclosure initiative and a dedicated threat intelligence team that monitors emerging risks around the clock. Together, these measures are intended to improve transparency while encouraging responsible reporting from independent security researchers.

AI Development Enters a New Regulatory Era

Anthropic’s latest announcement reflects a broader shift taking place across the artificial intelligence industry. Frontier AI models are becoming more capable, but they are also attracting greater regulatory attention as governments seek stronger oversight of technologies that could have far-reaching economic and security implications.

The successful restoration of Fable and Mythos demonstrates that regulatory compliance and technological progress do not have to work against each other. Instead, the experience has shown that stronger safeguards, independent security testing and closer collaboration between technology companies and government agencies can support responsible innovation while maintaining commercial momentum. Claude Sonnet 5 also reinforces Anthropic’s ambition to compete aggressively in the enterprise AI market by delivering improved coding performance, autonomous workflow capabilities and competitive pricing for organizations adopting AI at scale.

As businesses increasingly rely on advanced AI to automate software development, administrative processes and technical operations, confidence in safety and governance will become just as important as benchmark performance. Anthropic’s response to the recent export controls, combined with the launch of Claude Sonnet 5 and the return of Fable and Mythos, illustrates how the industry is evolving beyond simply building more powerful models. Success will increasingly depend on balancing innovation with accountability, transparency and security, ensuring that advanced AI systems can be deployed with greater confidence by enterprises, developers and public institutions alike.

For more details on Anthropic’s Claude Sonnet 5, visit:

https://www.anthropic.com/news/claude-sonnet-5

For the latest on new AI models debuting the market, keep on logging to Thailand AI News.