A safety push sparks infighting at OpenAI and Anthropic, AFR reports

A push on AI safety is fuelling internal friction at OpenAI and Anthropic, according to a report from The Australian Financial Review. The framing puts the two most closely watched frontier labs into a single story: publicly, safety commitments keep expanding; internally, disagreements over direction and pace are surfacing.

The item is reported narrative rather than a product or policy announcement. The available information covers only the headline claim, without the specifics of who disagreed, over what, or when, so it is best treated as a signal to track rather than a settled fact.

Why safety turns into internal friction has a structural explanation. Frontier labs must simultaneously iterate on capability to stay competitive and invest people and time in evaluation, alignment and outside audit, and those two demands tend to compete for the same researchers and the same attention.

The tension is not unique to one company. Over the past two years frontier labs have made extensive public commitments on safety staffing, pre-release evaluation and third-party testing. The more concrete the commitment, the more internal checkpoints it creates, and the more easily a technical question becomes an organisational one.

For outside observers the real question is not whether disagreement exists but whether it changes release cadence, evaluation standards or staff retention. Friction is not itself a risk if safety judgements are executed reliably through existing process; if they are not, it weakens how much weight outsiders give to public commitments.

What to watch next: whether OpenAI and Anthropic issue formal statements, whether safety teams' remits or reporting lines change publicly, and whether similar internal tension is reported at other labs.

Stretching the safety narrative from public statements into organisational life means the issue has entered the governance layer. For enterprise customers and regulators, judging a lab increasingly depends on how it handles disagreement internally, not only on what it says on stage.