EthicAI launches Selma as Australian firms seek AI control

EthicAI has launched Selma, a sovereign AI product for Australian organisations, after Anthropic suspended two AI models for customers worldwide following a US government order.

The Sydney consultancy says Selma is designed to reduce reliance on large US AI models by moving organisations to custom small language model architecture they own and operate. The system is tailored to each client's policies, workflows and Australian legal obligations.

Anthropic's suspension of Claude Fable 5 and Mythos 5 has heightened concern among some Australian organisations about the risks of relying on offshore AI infrastructure. EthicAI says Anthropic received a directive from the US Commerce Department and disabled both models globally within hours, leaving customers outside the US without access.

The episode has intensified a broader debate over who controls the infrastructure behind AI systems used in government, legal services and other regulated sectors. For organisations handling sensitive files or operating under statutory data obligations, the issue is not just performance or cost, but whether an external provider can cut off access without warning.

Selma is aimed at clients seeking to bring that infrastructure under their own control. EthicAI says its migration process builds a small language model around an organisation's internal documents and operational environment, with data kept inside the organisation and ownership of the resulting system retained by the client.

A second offering is aimed at organisations that do not want a full migration. In those cases, EthicAI says its ethical frameworks and guardrails can be added as a persistent layer across automated workflows and decision-making processes, either alongside Selma-based infrastructure or within an existing technology stack.

Market shift

EthicAI is entering the market as interest in smaller, locally run AI models grows. Large language models from major US providers dominated the first wave of corporate AI adoption, but smaller systems are attracting attention from organisations seeking tighter data control, lower operating costs and fewer dependencies on external providers.

The consultancy points to recent announcements by NVIDIA and Microsoft on tools for AI agents running on local hardware. It also cites NVIDIA Research findings that 40 to 70 per cent of everyday professional tasks can be handled by small language models without meaningful loss of effectiveness compared with frontier systems.

For Australian users, the argument has also taken on a geopolitical dimension. The shutdown of newly launched US models after a government order showed how quickly access can change when the underlying system is controlled in another jurisdiction.

"The writing was on the wall. This last week in AI news just confirmed it. The dependency on US frontier models is going to continue to become a problem for organisations trying to embed AI. The move to small language models is coming swiftly, and we believe it solves a lot of the issues we have been seeing around the ethical considerations for organisational use, including IP, environmental impact, and infrastructure," said Kara Bombell, co-founder of EthicAI.

Public sector focus

The consultancy says it has spent several years working with organisations facing strict constraints on the use of external systems for sensitive information. Those clients include legal aid services, child protection agencies, ombudsman offices, NDIS providers and government policy teams.

These sectors have adopted AI cautiously because of legal and regulatory obligations around data handling, records management and decision-making. A model hosted by an overseas provider can create additional risk when client files, case material or policy documents cannot be sent outside the organisation.

Selma launches with a signed pilot under way at a state government ombudsman office, giving EthicAI an early public sector reference point as it seeks to expand adoption among agencies and other regulated bodies.

The launch also coincides with a change to EthicAI's leadership team. Simon Brock has joined as chief product officer after senior roles across marketing and digital product at Publicis Groupe.

He will lead commercial strategy and product development for Selma as EthicAI looks to broaden its migration work. The company was co-founded by Bombell and Katriel Healy, whose advisory work in the Australian public sector, health and community sectors has shaped the product's focus.

The emergence of products such as Selma reflects a broader shift in the Australian AI market, as organisations look beyond early experiments with general-purpose US models towards systems more closely aligned with domestic regulatory settings and internal governance requirements.

For boards and regulators, a central issue is whether guardrails around automated decision-making operate consistently across workflows. EthicAI says that demand is creating room not only for full-stack AI systems owned by the client, but also for governance layers that can sit on top of existing tools.