AI-Assisted Cyber Operations: What Recent Reports Say

Recent reporting from ENISA, Anthropic and Prophet Security provides three different views of AI-assisted cyber operations. ENISA examines the European threat landscape, Anthropic describes disrupted misuse of AI in specific campaigns, and Prophet Security reports on AI-assisted investigation of customer alerts.

These reports are not directly comparable. They use different evidence, cover different populations and answer different questions. Taken together, however, they help separate a broad threat trend from concrete examples of offensive use and one defensive operating model.

The article is most useful for security leaders, identity teams, threat intelligence practitioners and SOC leaders deciding where AI changes their existing priorities—and where it does not.

Short answer

The reports suggest that AI is increasingly being used to accelerate familiar cyber activity rather than replace established attack techniques. ENISA describes AI as an enabler for speed, scale and adaptability across malicious operations. Anthropic provides case evidence of AI-supported reconnaissance, exploitation workflows, credential theft and data collection. Prophet Security focuses on a different question: how AI-supported investigation can help defenders identify identity and session-based attacks in customer telemetry. The available material supports treating AI as a factor in existing controls for identity, exposure management and detection, rather than as a separate threat category.

What the reports agree on

All three reports place identity and access at the center of practical cyber risk. ENISA describes phishing, credential compromise and unauthorised access as persistent concerns in the EU threat landscape. Anthropic documents campaigns involving compromised credentials, session tokens and API keys. Prophet Security reports that identity was the objective in roughly half of the malicious activity it confirmed during its reporting period.

They also point to AI being used around established stages of cyber activity. ENISA reports AI-supported phishing, reconnaissance, script development and post-exploitation. Anthropic describes operators using AI to coordinate reconnaissance, vulnerability research, exploitation and collection work. Prophet Security does not assess attacker use of AI in the same way, but it shows an AI SOC analyst using evidence from identity, endpoint and other telemetry to investigate alerts.

The reports also support a practical conclusion: controls need to account for what happens after initial access. ENISA describes credential compromise and abuse of trusted environments. Anthropic highlights credentials, API keys and sessions as both targets and operational resources. Prophet Security emphasizes that an already-authenticated session can require different containment and detection actions than a password attempt.

Where the reports differ

The reports have materially different scopes. ENISA is a broad assessment of incidents affecting EU Member States and EU-based organisations during 2025. It is designed to describe trends across cybercrime, state-nexus activity, hacktivism, vulnerabilities and sectors.

Anthropic is narrower and case-led. Its material describes observed misuse and disrupted activity involving AI systems, including offensive workflows, AI credential theft and attacks on AI-related systems. It offers concrete examples, but it is not a measurement of the whole threat landscape.

Prophet Security is different again: it is a vendor report based on alerts investigated in customer environments between May and July 2026. Its most relevant contribution here is defensive evidence about identity-focused attacks, session hijacking and investigation depth. It should not be read as a general measure of global attacker activity or as directly equivalent to ENISA's incident dataset.

This difference matters. ENISA helps readers understand the broader setting, Anthropic helps explain observed AI-assisted offensive workflows, and Prophet Security helps frame the detection and response implications. The reports complement one another only when their scope and methodology remain explicit.

Key Cyntari themes related to this topic

AI Security & AI Governance

ENISA and Anthropic both describe AI as a tool used in malicious activity and as an attack surface when AI systems, API keys, integrations or trusted access pathways are exposed.

Identity, Phishing & Access Security

The reports connect phishing, credentials, session tokens and API keys to initial access, persistence and misuse. Prophet Security particularly emphasizes the risk of already-authenticated sessions.

SOC, Detection & Response

Prophet Security focuses on AI-supported alert investigation, while the other reports provide attacker behaviours that detection and response teams may need to contextualize.

Threat Actors, Geopolitics & Intelligence

ENISA and Anthropic describe cybercrime, state-nexus and ideologically motivated activity, while cautioning that actor objectives and evidence vary across cases.

What this means for security teams

Do not treat AI-enabled cyber activity as separate from core security work. The available reports point back to familiar priorities: phishing resistance, credential and session protection, API-key handling, exposure reduction, logging and evidence-led investigation.

Identity teams can review whether their containment procedures address sessions and refresh tokens as well as passwords. Teams using AI services can treat AI credentials and integrations as production access paths, with the same care given to other sensitive credentials.

SOC leaders can use the reports to test whether investigations join identity, endpoint, cloud and historical account context. The reports do not establish a single best operating model, but they do show why isolated alert handling can miss the wider sequence of an intrusion.

Recommended reports to read

• ENISA — ENISA Threat Landscape 2026 — 2026 for a broad EU-focused assessment of cybercrime, state-nexus activity, social engineering, vulnerabilities and AI-related trends.

• Anthropic — Detecting and Countering Misuse of AI: September 2026 — 2026 for detailed examples of observed AI-assisted offensive workflows and AI ecosystem abuse.

• Prophet Security — Prophet Security Quarterly Threat Report Q3 2026 — 2026 for a customer-telemetry perspective on identity attacks, session hijacking and AI-assisted SOC investigation.