Predictions and guidance on AI-driven cyber risk and emerging threats in 2026
Commentary from Dark Reading and the Resilient Cyber newsletter highlights agentic AI and broader AI-enabled social engineering (including deepfakes) as growing enterprise attack-surface concerns heading into 2026, alongside continued emphasis on fundamentals like vulnerability management. A Dark Reading readership poll framed agentic AI as the most likely major security trend for 2026, reflecting expectations that increasingly autonomous systems will become attractive targets and/or tools for cybercrime.
A separate Dark Reading “Reporters’ Notebook” discussion urged security leaders to prioritize practical steps for 2026, including improving resilience against phishing/social engineering, accelerating patching, and preparing for quantum-era cryptography transitions. The Resilient Cyber newsletter echoed the “inflection point” theme for operationalizing AI security, citing model-provider discussions (e.g., OpenAI’s Cyber Preparedness Framework and Anthropic’s reporting on abuse) and arguing that defenders will need to adopt AI capabilities to keep pace with attackers, while acknowledging that guardrails can be bypassed and that AI-driven fraud (e.g., deepfake phishing) is already a near-term risk.
Get ahead of threats like this
Mallory correlates global threat intelligence with your attack surface — know if you’re exposed before adversaries strike.
How this story unfolded
10 events from the most recent confirmed update back to the earliest known activity.
CIO feature spotlights AI-agent threat modeling around autonomy
A CIO security feature discussed threat modeling for AI agents, emphasizing that increased autonomy expands the attack surface. The referenced content did not describe a specific incident, but it marked continued mainstream coverage of agentic AI security concerns.
Poll ranks deepfakes as next major social-engineering threat
The same poll found 29% of respondents expect deepfakes to become the primary social-engineering vector against high-value targets. The article pointed to examples including a reported $25 million Hong Kong fraud and North Korea's fake worker campaign.
Dark Reading poll finds agentic AI seen as top 2026 attack-surface risk
A Dark Reading readership poll found 48% of respondents expect agentic AI to become the leading attack vector for cybercriminals and nation-state actors by the end of 2026. The article tied this expectation to growing autonomy, privileged access, open-source components, and shadow AI adoption.
Industry panel calls for quantum-readiness and better patch prioritization
The same Jan. 29 panel urged organizations to begin preparing for quantum-resistant encryption by inventorying cryptographic use and engaging leadership on transition planning. Panelists also warned that defenders continue to ignore known vulnerabilities, citing ongoing exploitation of long-disclosed flaws.
Industry panel urges measured AI adoption and stronger phishing defenses
In a Jan. 29, 2026 Dark Reading discussion, security journalists argued organizations should avoid indiscriminate "AI everywhere" deployment and instead use AI selectively, such as for SOC triage. They also stressed reducing reliance on end users as the last line of defense by improving verification practices and stronger authentication.
Clawdbot case study highlights exposed agent deployments and skills abuse
A case study involving the open-source Clawdbot agent gateway described exposed deployments, abuse of third-party "skills," and a reported malicious VS Code extension that drops malware and weaponizes ScreenConnect. The example was presented as an early real-world illustration of agent ecosystem risk.
Research warns many agent 'skills' contain vulnerabilities
Research cited in the newsletter reported that a large share of agent "skills" contain security flaws, raising concerns about supply-chain-style abuse in agent ecosystems. The finding reinforced warnings that open agent integrations can expand enterprise attack surfaces.
UK NCSC issues guidance on vulnerability prioritization and prompt injection
The UK National Cyber Security Centre published guidance covering vulnerability prioritization and how to frame prompt-injection risk. The guidance was highlighted as part of growing official efforts to address AI-enabled and AI-targeted security issues.
Anthropic reports malicious use of its AI platform
Anthropic disclosed cases of malicious use of its platform, adding evidence that mainstream AI services are already being abused for harmful activity. This was cited as a concrete indicator that attacker use of AI is no longer hypothetical.
OpenAI signals move toward 'High' cyber preparedness level
OpenAI stated it is on a trajectory toward a "High" level in its Cyber Preparedness Framework, reflecting increased concern about the cyber capabilities of advanced AI systems. The newsletter cites this as a governance milestone in the broader shift from theoretical to operational AI security risk.
Related entities
Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.
Sources
4 references tracked. Mallory keeps watching after this page renders.
AIエージェント時代の脅威モデル入門──「自律性」が増やす攻撃面をどう捉えるか | CIO
cio.com
Open source2026: The Year Agentic AI Becomes the Attack-Surface Poster Child
darkreading.com
Open sourceFrom Quantum to AI Risks: Preparing for Cybersecurity's Future
darkreading.com
Open sourceResilient Cyber Newsletter #82 - by Chris Hughes
resilientcyber.io
Open sourceSee the full picture, correlated to your attack surface.
Map indicators from this story to your assets and identify affected systems in minutes.
Every observed campaign, victim, and pivot linked to actors named in this story.
Malware, exploits, and IOCs connected to the activity described here.
YARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Get matching new stories delivered to your team as they break — not the next morning.
Ask questions about this story and take action on the answers.
Related stories
Other stories Mallory is tracking that overlap on entities, sources, or topic labels.
Predictions and Impacts of AI on Cybersecurity in 2026
Artificial intelligence is expected to significantly shape the cybersecurity landscape in 2026, with experts forecasting both increased threats and the necessity for AI-driven defenses. Analysts predict that attackers will leverage agentic and autonomous AI to automate phishing, lateral movement, and exploit chains, reducing the need for human operators and increasing the scale and sophistication of attacks. Moody's 2026 outlook highlights the risk of adaptive malware, deepfakes, and model poisoning, warning that organizations adopting AI without robust safeguards will face heightened regulatory and operational risks. The rapid evolution of AI-powered threats is already evident, with more personalized phishing and faster, AI-assisted hacking becoming mainstream concerns. In response, cybersecurity professionals emphasize that AI-powered defense is no longer optional for modern organizations. Traditional security tools are increasingly ineffective against the speed and complexity of modern threats, such as ransomware and targeted phishing. AI-based cybersecurity solutions augment human expertise by analyzing vast datasets in real time, learning normal behavior patterns, and flagging anomalies with greater accuracy. This shift enables security teams to reduce false positives, detect subtle threats earlier, and prioritize responses more effectively, helping organizations mitigate regulatory penalties, operational downtime, and reputational harm in an environment where cyberattacks are more frequent and sophisticated than ever before.
Mar 21, 2026
Industry Commentary on Phishing and AI-Enabled Cyberattacks
Security commentary published in early 2026 highlights that **phishing remains highly effective** despite improved defensive tooling, largely because attackers exploit predictable human psychological triggers. One analysis frames phishing success as a three-stage process—*bait, hook, catch*—where adversaries research targets, deliver tailored lures, and then convert engagement (e.g., link clicks or credential entry) into compromise; it also cites CISA-reported prevalence of phishing in successful intrusions and notes that while overall phishing volume may fluctuate, financial impact can still rise. Separate reporting and analyst content focuses on **AI’s growing role in the attack chain** but stops short of confirming fully autonomous end-to-end attacks in the wild. An international AI safety report and related coverage describe AI systems assisting with tasks such as vulnerability scanning and malware development, and reference prior claims of **semi-autonomous** operations (with humans making key decisions), including reported abuse of an AI coding tool to support intrusions against dozens of high-profile organizations with limited success. A technology roundup aimed at CISOs ties these trends to increased 2026 security spending and prioritization of AI-enabled defenses, but it is primarily forward-looking guidance rather than incident-driven intelligence.
Mar 21, 2026
Mixed Cybersecurity Roundup: AI-Enabled Crypto Fraud, DDoS Campaigns, and 2026 Risk Predictions
Reporting in this set is not a single coherent incident; it is a **mixed roundup** dominated by (1) **AI-enabled cryptocurrency fraud** and (2) **DDoS activity and botnet trends**, alongside several forward-looking or non-incident items. Chainalysis-linked coverage describes industrialized crypto crime, including an estimate of **$17B in 2025 crypto-scam losses** and a sharp rise in **AI-driven impersonation/deepfake tactics**, with links to organized crime networks and forced-labor scam compounds in **Cambodia and Myanmar**; separate reporting notes a **$26.44M theft from the Ethereum-based Truebit protocol**, with Truebit urging users to avoid a **compromised smart contract** while investigations continue. In parallel, threat reporting highlights large-scale DDoS: Cloudflare’s mitigation of a **29.7 Tbps** burst attributed to the **AISURU** botnet-for-hire (plus a **14.1 Bpps** event and an estimated **1–4M** infected hosts), and a concentrated **NoName057(16)/DDoSia** campaign against the **UK** (1,812 attack entries targeting 86 domains/87 IPs, heavily hitting government and some critical infrastructure, with port **443** most targeted). Spamhaus also reports a **24% increase** in botnet C2 activity in 2H 2025, with **RATs** comprising a large share of top botnet-associated malware. Several items are **not incident-driven** and should be treated as lower-signal for operational response: SC Media and Security Boulevard pieces largely provide **2026 predictions/opinion** on *agentic AI*, **non-human identities (NHIs)**, and deepfakes as governance/identity risks; Dark Reading and CIO discuss **regulatory/compliance** and **IT leadership** challenges; TechTarget lists **2026 conferences**; and two Substack posts are general **news roundup/essay** content (one recounting lessons from Ukraine’s cyber conflict, including the Kyivstar destructive attack narrative). For CISOs, the actionable takeaways across the incident-focused items are: expect continued growth in **AI-assisted social engineering and deepfake fraud** impacting financial loss and brand trust; maintain smart-contract incident playbooks for rapid user guidance; and harden DDoS readiness (capacity planning, upstream mitigation, and monitoring) given both **record-scale botnet bursts** and **geopolitically motivated DDoS** targeting government and critical infrastructure.
Mar 21, 2026