Australian Signals Directorate warns of sudden service outages as AI threats rise

The Australian Signals Directorate issued its annual cyber threat report on Tuesday and said the public should expect sudden outages of important online services. The agency said organisations should prioritise system security over continuous availability. The agency said organisations and the public should brace for disruptions caused by defensive work or attacks instead of assuming steady availability. The agency chief and the defence minister said organisations must rank system security above continuous operation, a step that could force services offline during security work.

The agency said companies and essential infrastructure operators should use advanced AI models to defend against cyber threats. It also said organisations must match advanced defensive AI with faster upgrades to cybersecurity foundations. The agency can use Anthropic’s Mythos model, a powerful AI system kept from public release by Anthropic, a frontier lab that develops advanced foundation models, because it can identify unknown weaknesses in computer systems. Australian government agencies view access to frontier models and collaboration with leading AI labs as essential to stronger cybersecurity.

The agency said AI is generating more dangerous threats than before and will probably accelerate cyber threats going forward. Broader access to AI helps cybercriminals and state actors locate weaknesses and conduct attacks faster and more effectively. The report said AI provides advantages while also lowering obstacles and increasing the abilities of criminals and state-backed actors.

Malicious actors can apply AI to accelerate reconnaissance, identify weaknesses and run operations more efficiently. AI is pushing faster growth in how often and how costly cyber attacks are against Australian companies, governments and critical infrastructure. State-backed actors hunt for weaknesses in Australian networks, get inside and stay concealed until they decide to disrupt services or collect intelligence.

State-based actors made repeated attempts to breach government and private networks across Australia in the previous year. Actors in the Indo-Pacific may be running operations to get ready for possible future conflicts, including intelligence gathering or disruption.

The Australian Signals Directorate managed 1,500 cybersecurity incidents in the 2025-26 year, up 20% from the prior period. It notified over 4,400 businesses, governments and critical infrastructure operators about possible malicious cyber incidents, an annual increase of 80%.

The director general called legacy technology, including outdated or unsupported devices and systems, a major risk and an appealing target for hackers. The country’s spy chief urgently asked organisations to patch security holes in older technology. The agency pressed the country to improve its technology and cybersecurity capabilities.

Last week, newly revealed weaknesses in PHP, the programming language behind most websites, led governments and businesses to rush repairs. The agency told organisations to fix their systems, contact ASD for help and report suspicious activity. The coverage does not name specific tools, demand a single technical method or set deadlines for every organisation.

Abigail Bradshaw said on Monday that automated defences powered by AI are needed to stay ahead of attackers using AI. Richard Marles said on the report’s release that Australians must strengthen their cyber defences while making use of AI’s benefits. Tony Burke said the government is preparing Australia to handle evolving threats and opportunities from new technologies.

The agency said cybersecurity is now a core business risk and a leadership duty for every organisation, not just a technical issue. It said every organisation that owns a network now faces elevated threat levels.

The report said adding AI tools to security work might reveal weaknesses or suspicious activity and improve incident responses. Defensive tools that sort alerts, spot unusual behaviour and handle routine tasks can also allow adversaries to broaden intelligence collection and attack planning.