AI phishing pushes identity to the front line

By late 2025, AI-enabled social engineering and phishing attacks experienced an unprecedented surge, with KnowBe4 reporting a 449% increase in AI-powered vishing and a 67% rise in abuse of legitimate platforms. Attackers leveraged AI to rapidly generate thousands of highly personalized lures, often timed around seasonal events like tax deadlines and holidays, enabling scalable campaigns that bypassed traditional cybersecurity defenses by hijacking trusted platforms and exploiting contextual company details.

The effectiveness of AI-driven phishing was underscored in Q3 2025 when 90% of user interactions stemmed from emails personalized with company names and internal topics, especially those impersonating HR and IT departments. Coupled with the widespread use of branded landing pages (70%) and domain spoofing (66%), these tactics created convincing multi-modal attacks—including deepfake voice calls and interactive AI chatbots—that blurred the lines between genuine communication and malicious intent.

The emergence of novel attack vectors such as 'ChatOps Phishing,' where AI bots impersonate IT support in live chats to manipulate victims into compromising security, exemplifies the sophistication of AI-enabled social engineering. Additionally, platforms like the 'Quantum Route Redirect' automated phishing operations targeting Microsoft 365 users, incorporating advanced techniques like QR code-based 'quishing' to scale credential theft campaigns, reflecting a broader evolution in phishing tactics beyond traditional email.

Voice cloning technology supercharged vishing attacks by late 2025, enabling attackers to convincingly impersonate executives—as seen in the 2024 Ferrari CEO incident—and exploit inherent human trust in voice communication. This ‘lo-fi meets high-tech’ approach, combining simple call-to-action delivery with AI-powered persuasion, rendered traditional filtering ineffective and underscored the need for policy-driven verification processes. Meanwhile, attackers diversified their social engineering pools, recruiting female operatives to increase success rates in IT help desk impersonations, highlighting the adaptive and resilient nature of AI-enabled social engineering.

By late 2025, cybersecurity defense paradigms had decisively shifted from traditional perimeter-focused protections to identity-centric strategies, emphasizing integrated identity and SaaS posture assessments alongside continuous configuration audits to reduce privilege creep and credential exposure. This evolution is exemplified by defenses against sophisticated groups like Scattered Spider, where CISOs prioritize comprehensive identity visibility across human, machine, API, and vendor identities to create unified risk and access views, underscoring identity as the new security perimeter.

Throughout early 2026, attackers innovated rapidly to exploit identity verification mechanisms, notably through MFA phishing, OAuth abuse, and novel social engineering tactics such as the 'consent fix attack' that manipulates users into granting legitimate app permissions and leaking OAuth credentials. Campaigns like ShinyHunters’ targeted Okta SSO voice phishing, which registered over 150 custom domains to mimic MFA portals, highlight how adversaries bypass multi-factor authentication to gain persistent access, reflecting a broader trend where identity verification processes, rather than network perimeters, are the primary battleground.

High-profile incidents such as the $40 million Step Finance crypto theft and Iron Mountain’s brand impersonation phishing underscore the shift from attacking systems to exploiting trusted identities and communication channels. Attackers increasingly steal session tokens, cookies, and OAuth tokens to maintain stealthy, persistent access—often bypassing password resets and traditional detection methods—while leveraging trusted platforms and legitimate signed software to blend into normal corporate traffic, as seen in phishing campaigns exploiting Zoom and Microsoft Teams invitations.

By mid-2026, identity and trust exploitation had industrialized, compressing attacker dwell times from days to mere minutes and expanding attack surfaces to include OAuth tokens, AI plugins, and third-party integrations. This rapid evolution, coupled with widespread credential leaks—such as a 47-fold spike in dark web alerts impacting manufacturing—and the commoditization of infostealer malware as subscription services, has forced organizations to adopt zero trust frameworks and continuous validation of trust relationships. As Devon Ackerman aptly states, 'attackers aren't breaking in, they're logging in, then operating inside tools you've already approved,' making trust itself the central challenge in modern cybersecurity.

By 2025, cybercriminals had dramatically escalated their tactics, with KnowBe4 reporting a 449% surge in AI-powered vishing attacks and a 67% increase in the abuse of legitimate platforms to bypass traditional defenses. Groups like Scattered Spider exploited high-profile supply chain breaches at retailers such as M&S and Harrods, leveraging brand impersonation to fuel widespread phishing campaigns that caused hundreds of millions in damages, signaling a shift toward multi-modal, highly targeted attacks.

The evolution from mass phishing to precision-targeted campaigns became evident as attackers employed AI to automate intelligence gathering, constructing detailed organizational persona graphs from LinkedIn and news sources to mimic individual communication styles. This enabled sophisticated multi-modal phishing across platforms like Slack, Teams, and ServiceNow, often layering AI-generated deepfake voicemails and interactive chatbots to orchestrate context-aware social engineering that manipulates victims through coordinated vishing calls paired with MFA push notifications, exemplifying a move from brute force to psychological manipulation.

The rise of AI voice cloning profoundly transformed vishing, as demonstrated in a 2024 attack on Ferrari’s CEO, where scammers convincingly impersonated high-profile individuals. Despite the dramatic 442% increase in vishing incidents reported by CrowdStrike, many organizations remain vulnerable due to security training gaps that overlook phone-based phishing. At DEFCON 2024, AI chatbots outperformed human social engineers in adaptability and effectiveness, underscoring the urgent need for enhanced defenses against AI-driven voice phishing.

Targeted campaigns like ShinyHunters’ 2026 voice phishing assault on Okta SSO users illustrate attackers’ strategic focus on high-value sectors, registering approximately 150 custom domains to impersonate help desks and steal MFA codes. By registering their own MFA devices, attackers maintain persistent network access, enabling lateral movement and data exfiltration for extortion. This exploitation of single sign-on services as critical single points of failure highlights the increasing sophistication and persistence of identity-centric attacks.

By early 2026, the professionalization of social engineering was evident as the Scattered Lapsis Hunters alliance recruited women for vishing attacks, offering upfront payments up to $1,000 per call and providing scripted dialogues to boost success rates. This supergroup, comprising Lapsis, Scattered Spider, and Shiny Hunters, has compromised major global corporations, stealing over 1.5 billion records, reflecting a highly incentivized and organized approach to exploiting human trust in voice-based attacks.

Attackers have expanded beyond digital channels, employing physical mail phishing campaigns that leverage data breaches like Ledger’s 2020 exposure of 270,000 customer records to deliver personalized, official-looking letters containing QR codes leading to credential theft. This multi-modal escalation, amplified by AI’s ability to generate tailored content and deepfake impersonations at scale, bypasses digital defenses entirely, demanding new strategies that address both physical and digital trust exploitation.

Sophisticated phishing attacks increasingly exploit trusted relationships and technical misconfigurations to bypass security measures. For instance, attackers have used compromised Google Workspace accounts to send convincing emails mimicking regular communication, employing man-in-the-middle techniques to capture credentials and two-factor codes, often remaining undetected for months. Such campaigns, potentially affecting thousands of accounts simultaneously, underscore the systemic risks posed by identity and trust exploitation in high-value sectors.

By early 2026, trust relationships within organizations and cloud ecosystems emerged as the primary attack vector, with attackers increasingly exploiting these connections rather than traditional system vulnerabilities. Incidents like the ShinyHunters' OAuth token phishing campaigns targeting over 100 organizations and Fortinet’s FortiCloud patch issues illustrate how attackers weaponize trusted third-party integrations and cloud environments. This shift underscores the growing complexity of managing and continuously validating trust in multifaceted authentication flows, as exemplified by the novel ConsentFix attack that manipulates users into granting OAuth permissions, effectively turning legitimate applications into conduits for credential capture.

The erosion of trust is starkly evident in attacks targeting human elements within organizations, where compromised executive devices and shared mailboxes serve as gateways to critical assets. The $40 million crypto theft from Step Finance, achieved by hacking the executive holding the keys rather than the blockchain itself, alongside healthcare payroll heists leveraging shared mailbox credentials and help desk impersonations, reveal how attackers bypass perimeter defenses by impersonating trusted insiders. These cases highlight that when attackers target trusted identities, traditional security tools often fail to detect breaches, emphasizing identity as the new cybersecurity perimeter.

Attackers have refined their tactics to exploit trust not only within organizations but also across communication channels and hybrid environments, leveraging curated datasets and legitimate platforms to conduct sophisticated phishing and token-based compromises. The breach at Aura exposed 900,000 marketing contacts enabling targeted campaigns, while Nordstrom’s email system was hijacked to send fraudulent cryptocurrency scams from trusted domains. Furthermore, the rise of AI-assisted identity chain attacks, which use compromised credentials to escalate privileges across hybrid infrastructures, demonstrates how attackers capitalize on existing trust relationships, making continuous validation of trust more critical than ever.

As organizations adopt AI, cloud-native development, and interconnected ecosystems, the challenge of protecting trust relationships has escalated to rival the importance of securing infrastructure itself. Attackers now target critical trusted systems such as VPNs, SD-WAN controllers, and AI platforms, exploiting the inherent reliance on these technologies that were never designed for sustained adversarial pressure. Security experts warn that the core vulnerability lies not in traditional security controls but in the lack of visibility into how trusted tools like OAuth apps, cloud APIs, and AI assistants are weaponized—illustrated by attacks exfiltrating millions of files via Microsoft Graph API without triggering alerts. Consequently, continuous trust validation, zero trust principles, and expanded governance over SaaS integrations and third-party access have become indispensable to mitigating this pervasive erosion of trust.

By late 2025, defensive strategies against AI-driven phishing attacks had decisively moved away from static, signature-based detection toward a layered defense model integrating people, processes, and technology. Organizations embraced Just-in-Time Micro-Drills powered by AI to deliver realistic, frequent training, coupled with clear policies such as mandatory out-of-band verification for critical requests and streamlined phishing reporting mechanisms, transforming employees into active sensors within the security framework. Technological advancements included enforcing strict email authentication protocols (SPF/DKIM/DMARC), limiting OAuth app approvals, and adopting phish-resistant MFA methods like FIDO2 and passkeys, all aimed at countering the infinite variations of AI-generated phishing lures.

By November 2025, the cybersecurity focus had expanded beyond perimeter defenses to embrace integrated identity-centric frameworks that unify human, machine, AI, and vendor identities across cloud, SaaS, and on-premises environments. This comprehensive identity visibility enabled continuous posture hardening through configuration audits, privilege reduction, and threat-informed risk modeling, while behavior-based threat detection leveraged runtime anomaly detection and cross-system telemetry correlation, exemplified by Permiso’s 1,500+ advanced detection rules. This evolution recognized identity as the new security perimeter, where compromising a trusted identity could bypass traditional controls, necessitating layered defenses that monitor and respond to identity and credential threats in real time.

In early 2026, high-profile incidents like Step Finance’s $40 million crypto theft underscored the critical need for multi-sig hardware wallets and device separation to protect executive-level transactions, while persistent risks from brand impersonation and social engineering scams drove adoption of organizational policies enforcing invoice payments exclusively through authenticated vendor portals. Concurrently, the rise of AI-enhanced voice cloning attacks revealed the limitations of technical filtering alone, prompting companies to implement policy-driven verification processes that empower employees to independently verify identities—even at the risk of offending executives—highlighting trust as both a vulnerability and a defensive focal point.

By mid-2026, identity-related threats surged dramatically, with reports from Red Canary noting an 850% year-over-year increase in identity threat detections and SpyCloud revealing over 65 billion distinct identity records in their data lakes, including non-human identities. This explosion of compromised credentials—totaling over 414 billion—enabled attackers to bypass exploits by logging in as legitimate users, rendering traditional OTP-based MFA increasingly ineffective against real-time interception attacks documented by Google’s Threat Intelligence Group. Consequently, organizations accelerated adoption of phishing-resistant authentication methods such as FIDO2, passkeys, and hardware-backed models, while enhancing detection capabilities through sophisticated behavioral monitoring, zero trust frameworks, and continuous identity verification, firmly establishing identity as the foundational battleground of modern cybersecurity.

Even the best technical controls falter when attackers exploit human trust—prompting a shift to policy-driven verification and employee empowerment as frontline defenses against AI-powered deception.