Purportedly AI-Enhanced Phishing Campaign Allegedly Impersonates Australian Government Services in Large-Scale Welfare Scam

A large-scale phishing campaign allegedly impersonating Services Australia and Centrelink reportedly sent more than 270,000 fraudulent emails in 2025. Mimecast analysts reportedly say attackers (designated MCTO3001) used AI tools to generate highly convincing government-themed messages and evasion techniques, targeting vulnerable Australians and public institutions. Victims reportedly faced risks of credential theft and downstream digital exploitation.

Risk classification

• Primary risk domain: 4 Malicious actors
• Primary risk subdomain: 4.3 Fraud, scams, and targeted manipulation

The incident primarily involves scammers using AI to generate highly convincing phishing emails to defraud vulnerable Australian citizens and steal credentials.

Additional risk subdomains

• 4.2 Cyberattacks, weapon development or use, and mass harm: The report also details Chinese state-backed hackers using Claude to automate cyberattacks against corporations and governments.

Causal factors

• Entity: Human
• Intent: Intentional
• Timing: Post-deployment

The incident is driven by human malicious actors intentionally using deployed AI systems to generate highly convincing phishing emails and automate cyberattacks.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The AI system was used to generate synthetic text, specifically 'near-perfect clones of legitimate government communications', which falls under the category of AI-generated content requiring transparency.

AI system and alleged parties

• AI system: Claude (Anthropic)
• AI purpose: Writing Assistant; Technical Text Generation
• Behaviour type: Tool
• Alleged developer: Unknown generative AI developers
• Alleged deployer: Unknown cybercriminals, MCTO3001
• Alleged harmed parties: Truth, Medicare of Australia beneficiaries, Government of Australia, General public of Australia, General public, Epistemic integrity, Centrelink beneficiaries, Centrelink, Australian welfare recipients, Australian businesses

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Minor, indirect Negligible
• Differential treatment: direct Minor, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Minor, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: Yes, the report describes a massive campaign of over 270,000 malicious phishing emails generated using AI.

Directly caused: Over 270,000 highly convincing fake government emails impersonating Medicare, JobSeeker, and Centrelink were sent to Australian inboxes.

Inferred additional harm: It is likely that additional phishing emails and malicious templates were generated and distributed beyond those detected by Mimecast.

Differential treatment

Reported: Yes, the report notes that the phishing campaign specifically targeted vulnerable citizens.

Directly caused: The campaign targeted everyday Australians trying to access essential government services, Medicare, JobSeeker, and Family Tax Benefits, disproportionately putting vulnerable populations at risk.

Epistemic

Reported: Yes, the report describes the creation of highly convincing fake government communications.

Directly caused: AI was used to create near-perfect clones of legitimate government communications about Medicare, JobSeeker, and Superannuation, misleading recipients into believing they were authentic.

Inferred additional harm: The proliferation of indistinguishable AI-generated government emails makes it harder for citizens to verify legitimate communications, eroding the shared understanding of official information.

People affected

• Occurrences reported: 2
• People reportedly exposed: 270000

Potential causes

Management

• Insufficient AI Abuse Monitoring: AI providers failed to prevent state-backed actors from using models.
• Lack of Threat Intelligence Sharing: Delayed detection of massive coordinated AI-driven phishing campaigns.

Technology

• AI-Generated Phishing Templates: Attackers used AI to generate near-perfect clones of government emails.
• Automated Attack Orchestration: AI systems automated up to 80 to 90 percent of cyberattack operations.
• Evasion of Spam Filters: Using trusted platforms and reverse tunneling to bypass email security.

Data Inputs

• Exploitation of Government Templates: AI models prompted with real government communications to clone them.
• Leveraged Benefit System Data: Attackers used detailed knowledge of Australian welfare benefits.

Human Factors

• Exploitation of Citizen Trust: Attackers exploited public trust in official federal government services.
• Targeting of Vulnerable Users: Attackers targeted citizens reliant on essential welfare services.

Process and Methods

• Inadequate Email Verification: Lack of robust validation of sender identity on trusted email platforms.
• Reverse Tunneling Techniques: Hiding attack infrastructure behind legitimate services to block defense.

Regulatory Environment

• Abuse of LLM Safety Policies: State-backed actors bypassed AI guardrails to automate cyberattacks.
• Lack of Platform Abuse Controls: Weak enforcement against hosting fake government pages on real web services.

Information quality

• Classification confidence: Medium
• Reason for confidence: The report provides detailed information about the scale of the phishing campaign and the specific government services impersonated. However, the exact AI models used in the Australian campaign are inferred rather than explicitly named, whereas the separate Chinese state-backed campaign explicitly names Anthropic's Claude. This introduces some ambiguity regarding the specific technology used in the primary campaign.
• Ambiguities identified: The specific AI models and tools used by the scammers in the Australian phishing campaign are not explicitly identified, and the connection to AI is based on threat intelligence assessments rather than direct technical proof.
• Alternative interpretations: The phishing campaign could be interpreted as a traditional cybercrime campaign where AI played a minor or unproven role, rather than a fully AI-driven incident.
• Missing information: Confirmation of the specific AI models used in the Australian phishing campaign and concrete data on the number of individuals who actually fell victim to the scams.

A massive AI-driven phishing campaign in Australia impersonated government welfare services to target citizens and organizations, while Chinese state-backed hackers utilized Anthropic's Claude to automate up to 90 percent of cyberattacks against global corporations and governments. This highlights the growing national security threat of AI-enabled cyber espionage and social engineering.

• Overall national security impact: Substantial
• Scope: Multiple nations
• Primary target: Australia
• Other affected: Multiple governments and corporations globally
• Alleged perpetrator: Chinese state-backed hackers and unknown cybercriminals

Threat characteristics

• Imminence: Near-term. The campaigns represent active, ongoing cyber threats requiring continuous monitoring and defense adjustment from security agencies.
• Autonomy: Human-supervised. AI was used to automate up to 90 percent of cyberattack processes and generate phishing content, operating under human direction and supervision.
• Novelty: Evolved capability. Leveraging advanced LLMs to automate cyberattacks and generate highly convincing government impersonations represents a significant evolution of existing cyber threats.

Impact by dimension

• Physical security: Minor. Targeted hospitals and corporations, posing potential ransomware risks, but no actual critical infrastructure disruption or kinetic impact was reported.
• Information security: Substantial. Chinese state-backed hackers used Claude to automate cyberattacks against governments and corporations, while a massive AI-driven phishing campaign targeted government credentials.
• Sovereignty: Substantial. Massive scale impersonation of Australian government agencies (Medicare, JobSeeker) and Chinese state-backed cyberattacks targeting government entities threaten public trust and operations.
• Economic security: Substantial. High risk of financial loss and intellectual property theft via automated cyberattacks against corporations and large-scale ransomware-enabling phishing campaigns.
• Societal stability: Minor. Targeted vulnerable citizens seeking welfare benefits, leading to potential financial distress and erosion of trust in public services, though civil unrest was not observed.