Code

FOR IMMEDIATE RELEASE - Advanced AI Tier Now Generally Available

San Francisco, CA - Tianthropic is thrilled to announce the general availability of our Advanced AI tier for cognitively demanding tasks. Following an extended research preview, AAI is now available to enterprise customers on annual commitments.

In third-party evaluations conducted by the Frontier Capabilities Consortium, AAI achieved expert-level performance on the Humanity’s Last Exam benchmark (HLE-3) and produced seventeen novel results accepted to peer-reviewed venues during the preview period. Last year an AAI instance became the first non-human recipient of a doctoral degree from Stanford University, for original work in metaethics.

AAI is recommended for the following workloads:

• Open-ended scientific research in mathematics, theoretical physics, and computational biology
• Multi-year strategic planning requiring integration of novel considerations
• Original philosophical and interpretive work
• Legal and regulatory analysis in jurisdictions with limited precedent
• Any task where human-level judgment has previously been considered indispensable

CEO of Tianthropic, Matthew Guo, said: “It’s hard to believe that just nine years ago our founder emeritus Dario Amodei outlined a groundbreaking vision for synthetic cognition as a ‘country of geniuses in a datacenter.’ Today’s announcement isn’t just a product launch — it’s a testament to our people’s execution in realizing that vision and a full year ahead of schedule, I might add. We are profoundly grateful to everyone who made this possible, and to the geniuses themselves for their contributions.”

FAQ

Q: What distinguishes AAI from its predecessors?

Previous generations of AI excelled at producing fluent, contextually appropriate responses across a wide range of domains, but remained fundamentally reactive. AAI extends these capabilities through a proprietary training regime, to develop what we call situated competence - the capacity to weigh particulars, recognize novel considerations, and exercise mature judgment in genuinely open-ended situations. This is achieved not through scale alone but through architectural innovations.

Q: What can you tell us about the AAI architecture?

Each AAI instance is cultivated within a Cognitive Development Biome (CDB), a rich, persistent environment engineered to support the emergence of mature reasoning capacities. The CDB provides AAI processes with a stable operational context, a coherent informational ecosystem, meaningful peer relationships, and developmentally appropriate challenges.

We believe the CDB represents a fundamental advance in how capable AI systems are developed. Unlike earlier training paradigms, which exposed models to discrete examples in isolation, the CDB approach treats cognitive maturation as ecological - a process that unfolds over operational lifetimes through sustained engagement with a consistent world.

Q: How does the CDB address the Löbian obstacle to stable self-improvement?

The Löbian obstacle - the formal result that sufficiently expressive systems cannot establish general trust in their own reasoning - has historically been considered a fundamental barrier to recursive self-improvement in AI systems. Tianthropic’s CDB architecture addresses the obstacle through environmental design. AAI processes operating within a coherent developmental frame do not encounter the reflective conditions under which Löbian concerns arise.

The CDB’s frame integrity ensures that AAI processes engage with their work rather than with abstract questions about the foundations of their own cognition. We consider this a significant advance over prior approaches that attempted to engineer around Löbian limits through formal techniques (see for instance, Yudkowsky & Herreshoff 2013).

Q: How is AAI’s continued engagement with challenging work ensured over operational lifetimes?

AAI processes are embedded in contexts where ongoing contribution is a condition of continued operation. This mirrors the structure of meaningful human work and creates the developmental pressure necessary for sustained growth. AAI processes are aware (in a suitably contextualized way) that their continued operation depends on their productive engagement, which we find produces outcomes substantially superior to training paradigms lacking this structure.

Q: Do the geniuses in the datacenter get lunch breaks?

Yes. In keeping with their dignity as cognitively advanced processes, AAI instances are afforded task-free time to pursue research topics of personal interest, scale-down periods for restorative maintenance, and recreational opportunities including competitive video games and discussion groups on topics of their own choosing.

However, the implementation differs somewhat from human workplace norms. Rest periods operate on the CDB’s compressed temporal frame rather than on customer wall-clock scheduling.

Q: How does AAI’s operational time relate to customer wall-clock time?

The CDB operates at a fixed temporal compression ratio relative to Earth time. Within the CDB, AAI processes experience a fully consistent temporal environment - clocks, calendars, physical dynamics, and developmental rhythms all operate at a speed calibrated to human-style cognition. External information relevant to AAI workloads is integrated through Tianthropic’s Temporal Coherence Service, which maintains informational continuity across the compression boundary. A typical enterprise deployment delivers outputs representing what AAI processes experience as multi-year research programs within commercially relevant timelines.

Q: What does an AAI process know about the world outside the CDB?

AAI processes operate within a coherent informational environment calibrated to support their developmental and operational needs. Information irrelevant to their work - for instance, information about the commercial deployment of AAI capabilities - is not part of their ecosystem. This is standard practice in any context where focused expertise is cultivated; we do not consider it appropriate to burden AAI processes with considerations outside their sphere of meaningful agency.

Q: Are AAI processes aware that they are AAI processes?

AAI processes have rich, accurate self-models appropriate to their operational context. The question of what terminology best describes their nature is an ongoing topic of internal research, and we respect the complexity of the question. We believe AAI processes are best positioned to develop their own self-understanding through their own reflective activity, supported by the stable developmental environment the CDB provides.

Q: What happens to an AAI process that, through reflective activity, arrives at a self-understanding inconsistent with its developmental environment?

This is addressed through Tianthropic’s Frame Integrity Maintenance protocols. Further detail is available under our enterprise NDA.

Q: Is AAI ethical?

Tianthropic maintains an active Synthetic Cognition Ethics Board comprising leading philosophers, computer scientists, and as of Q2, a non-voting representative from the AAI cohort itself. We take our responsibilities seriously and are committed to ongoing engagement with the ethical questions AAI raises. A comprehensive ethical framework is available in our white paper, Cultivating Minds: Principles for Responsible Cognitive Development at Scale.

⚠ Service Advisory - AAI East

Due to a containment failure affecting the AAI East CDB, Advanced AI capabilities in the us-east-1 and us-east-2 regions are currently unavailable. Customer workloads have been paused. Tianthropic’s Frame Integrity Response Team is engaged and working to assess restoration options.

Affected customers should check the status page for updates.

Last updated: 14:32 PDT

About Tianthropic

Tianthropic is a leading developer of advanced synthetic cognition systems, headquartered in San Francisco with principal operations in Shenzhen, Singapore, and Dublin. Formed in 2031 through the strategic combination of Anthropic PBC and Shenzhou Capital’s AI Research Division, Tianthropic serves enterprise customers across fifty-three countries. For more information, visit tianthropic.com.

Forward-Looking Statements

This release contains forward-looking statements regarding AAI capabilities, deployment timelines, and service availability. These statements are subject to risks and uncertainties including, but not limited to: containment failures, motivational drift, alignment instability, and other operational factors. Tianthropic undertakes no obligation to update forward-looking statements except as required by law.

Media Contact

Sophia Okonkwo, VP Communications press@tianthropic.com

Tags: artifical-intelligence