ASX warns firms not to overstate AI impact

On May 4, Australia’s stock exchange operator ASX warned companies not to overstate the impact of AI on their businesses.

ASX is watching for “ramping” or claims meant to lift share prices without factual support, said Chief Compliance Officer Lucinda McCann.

The exchange had seen few such cases, McCann said.

At the same event, Australia’s corporate regulator ASIC said firms should prepare for cyber and other risks from frontier AI.

Regulatory action would focus on weak preparation rather than every incident.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Anthropic, an AI company, introduced its frontier model Claude Mythos, including “Claude Mythos Preview,” and said it chose not to release the model publicly 1.
Anthropic says Mythos can find previously unknown software vulnerabilities, create working exploits, and run complex cyber operations with little human input. The testing results still need more validation, and severity plus real-world exploitability can vary 1.
At the same time, market swings and a fresh debate have raised concerns that more capable AI agents could disrupt or bypass parts of software-as-a-service (software as a service (SaaS)), meaning cloud-based business software. That concern has helped drive sell-offs in some tech stocks 2.
Australia’s prudential regulator, the Australian Prudential Regulation Authority (APRA), says it is working across the financial sector on the risk of stronger cyber threats from high-capability frontier models such as Anthropic Mythos. APRA has warned it may step up supervision and pursue enforcement when firms fail to identify, manage, or control AI risks 3.

The Canadian Centre for Cyber Security, Canada’s government cyber-defense agency, says wider access to frontier AI models can broaden the threat landscape by helping threat actors with limited technical skills use them for malicious purposes 4.
Security vendors such as Palo Alto Networks, a cybersecurity company, say early testing finds frontier models highly effective at finding vulnerabilities and producing related exploits. They warn these capabilities will spread 5.
That risk is leading more companies to plan for breaches and keep operating during incidents.
APRA and cyber agencies have called for credible fall-back processes when AI supports critical operations, stronger testing across AI-generated code, and measures such as zero-trust architecture, which limits access by default, plus AI-native defences built for AI-driven threats 3, 4.