This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
A large-scale phishing campaign (MCTO3001) impersonating Australian government services has sent over 270,000 fraudulent emails, leveraging AI to create highly convincing messages and evade security filters. The campaign targets vulnerable individuals and various institutions, posing risks of credential theft and ransomware. Additionally, the report highlights separate evidence from Anthropic regarding Chinese state-backed hackers using the Claude model to automate cyberattacks against corporations and governments.
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.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to gain a personal advantage over others such as through cheating, fraud, scams, blackmail or targeted manipulation of beliefs or behavior. Examples include AI-facilitated plagiarism for research or education, impersonating a trusted or fake individual for illegitimate financial benefit, or creating humiliating or sexual imagery.
Also classified as
The report also details Chinese state-backed hackers using Claude to automate cyberattacks against corporations and governments.
The incident primarily involves scammers using AI to generate highly convincing phishing emails to defraud vulnerable Australian citizens and steal credentials.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The incident is driven by human malicious actors intentionally using deployed AI systems to generate highly convincing phishing emails and automate cyberattacks.
Added categories peak at 3: Substantial (Epistemic Harm, Inferred)
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.
The AI systems in this incident are primarily used as tools to generate highly convincing phishing emails and clone legitimate government communications. Additionally, Claude was used as a tool to automate aspects of cyberattacks, though the report notes a high level of automation with minimal human interaction.
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.
The AI was used as a writing assistant to draft highly convincing, near-perfect clones of legitimate government communications and phishing emails.
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.
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.
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.
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.
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.
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.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
