SpaceX claims to have secured $40 billion for AI chips as Deloitte faces a repayment demand over a flawed report

Other stories

ACCOUNTABILITY · CONSULTING · PUBLIC CONTRACTS

Investigation published on 25 October 2026

Australia seeks repayment of Deloitte’s fees over errors in an AI-assisted report

According to The Guardian, authorities in Western Australia have requested expedited repayment of the fees paid to Deloitte after errors were found in a study on the possible merger of three universities.

The government had paid A$166,000 for the report, which contained AI-generated fabricated bibliographic references and substantial discrepancies in its financial calculations.

In one scenario, costs of A$514 million and A$145 million were presented as adding up to A$399 million, when the correct total is A$659 million. The discrepancy is A$260 million. [3]

In another scenario, items totalling A$512 million were recorded as A$823 million, a difference of A$311 million. [3]

As The Guardian reports, Deloitte disclosed in the report that it had used generative AI to gather research, synthesise information and proofread the text. However, it has not been established that AI caused the specific arithmetic errors.

The authorities say they have been advised that the errors do not affect the underlying economic model or its findings. Deloitte is to review and reissue the document. The reporting describes a repayment demand, rather than a court ruling on liability. [3]

Why it matters. Using artificial intelligence does not remove the obligation to check professional work.

When a report supports public decisions, errors can raise questions about contractual performance, repayment of fees and professional responsibility.

The story shows that quality controls cannot stop at detecting nonexistent references. They must also check figures, calculations, assumptions and internal consistency.

GEOPOLITICS · TECHNOLOGICAL SOVEREIGNTY

Investigation published on 25 October 2026

Countries try to avoid choosing between American and Chinese artificial intelligence

Associated Press published an investigation on Sunday into how technological competition between the United States and China is shaping other countries’ decisions.

Washington promotes the Pax Silica initiative, while Beijing backs the World AI Cooperation Organization.

Kazakhstan is a telling example.

The country joined US cooperation initiatives on AI supply chains in June and subsequently joined the Chinese-backed organisation in July.

Other states are also trying to retain access to both ecosystems.

US authorities have warned that substantial involvement in competing technological initiatives may be incompatible with certain Pax Silica commitments.

China, meanwhile, highlights the openness of its models and the possibility of international technology relationships that are less dependent on the United States. [4]

The story published on 25 October is a journalistic investigation of an existing trend, rather than the announcement of a new international agreement that Sunday.

Why it matters. Choosing AI suppliers can determine access to processors, cloud infrastructure, data and international supply chains.

For European businesses, this reinforces the importance of avoiding unnecessary dependencies and developing realistic strategies for technological autonomy.

GOVERNANCE · ETHICS · SOCIETY

Investigation published on 25 October 2026

AI companies consult philosophers and religious representatives to define their models’ values

Associated Press published an investigation on Sunday into how AI developers are trying to incorporate different moral perspectives into their systems’ behaviour.

Anthropic had previously consulted scholars, clergy, philosophers and ethicists from more than 15 religious and cross-cultural groups, according to its announcement of 19 May. [5]

Some participants see these conversations as rigorous and a meaningful effort to incorporate different perspectives.

Other academics warn against treating religions and cultural traditions as interchangeable sets of moral rules.

They argue that combining different perspectives into a single system can erase nuances and favour a homogeneous interpretation of values that actually contain significant contradictions. [5]

The newly published investigation examines earlier dialogues. It does not announce a new model or the adoption of a binding ethical standard.

Why it matters. As AI systems gain autonomy, the principles guiding their decisions become more important.

Not every moral conflict can be resolved through optimisation calculations.

Organisations will still need to define who sets priorities, what limits apply to automated actions and who is accountable when decisions cause harm.