Press Portfolio

On 12 June 2026, the US Commerce Department ordered Anthropic to suspend access to its Claude Fable 5 and Mythos 5 models, citing national security. Anthropic received the directive at 5:21pm Eastern time and complied within hours, disabling both models for every customer globally. The model had launched three days earlier. Australian organisations had no warning and no recourse.

It is the starkest demonstration yet of what dependency on US frontier AI infrastructure means in practice: the control sits elsewhere, and it can be exercised overnight.

EthicAI, a Sydney-based responsible AI consultancy, is today launching Selma, a sovereign AI solution that moves organisations off that dependency onto a custom SLM architecture they own and operate, fine-tuned to their policies, protocols, and Australian legal context.

EthicAI works with each client to transition them to a small language model built specifically for their operational environment: their documents, their workflows, the regulatory and legal obligations that govern their sector in Australia. At the end of that process, the organisation owns its AI infrastructure outright, with no data leaving the organisation, and no third party who can issue a directive affecting it.

A second pathway serves organisations that have existing infrastructure and are not seeking a full migration. EthicAI's ethical frameworks and guardrails can be embedded as a persistent layer across automated workflows and decision-making processes, available as a subscription integrated with Selma-built infrastructure, or added to an organisation's existing technology stack. As automated decision-making scales across the public sector and regulated industries, the question of whether ethical guardrails are functioning consistently is one that boards and regulators are beginning to press directly.

The Fable 5 shutdown coincides with a broader signal from the market. At COMPUTEX 2026 and Microsoft Build this month, NVIDIA and Microsoft jointly announced tooling for AI agents running entirely on local hardware. NVIDIA Research has found that between 40 and 70 percent of everyday professional tasks can be handled by small language models without meaningful loss of effectiveness, at substantially lower cost and environmental footprint than frontier systems. The infrastructure argument for SLMs has been building for some time. The geopolitical argument arrived this week.

EthicAI has spent several years working with organisations that had the most to lose from getting AI architecture wrong: legal aid services, child protection agencies, ombudsman offices, NDIS providers, and government policy teams operating under statutory data obligations that make sending client files to any external system legally untenable. Selma launches with a signed pilot already running at a state government ombudsman office.

The launch also marks the expansion of the EthicAI leadership team. Simon Brock joins as Chief Product Officer, with years of senior leadership across marketing and digital product at Publicis Groupe. Brock will lead the commercial strategy and product development needed to take Selma's migration work to scale.

EthicAI was co-founded by Kara Bombell and Katriel Healy, whose advisory work across the Australian public sector, health, and community sectors forms both the distribution network and the domain knowledge that Selma is built on.