The security landscape has shifted dramatically in the first quarter of 2026. What began as experimental use of large language models by threat actors in 2024 has matured into fully weaponized attack tooling — and the volume of incidents we're seeing reflects it. Reported business email compromise (BEC) attempts are up 340% year-over-year, and the quality of phishing lures is nearly indistinguishable from legitimate correspondence.
If your firm's security posture hasn't been reassessed since 2024, it's almost certainly behind the curve. Here's what's changed for advisory firms specifically, and what your CCO and IT partner should be doing about it.
What's actually new
The scary part isn't that attackers are using AI — they've had access to the same models your marketing team has. The scary part is how they're using it.
1. Hyper-personalized phishing at scale
Previously, a spear-phishing campaign targeting a firm's CCO or principal took hours of manual reconnaissance. Now, an automated pipeline scrapes LinkedIn, your firm ADV, your custodian, and any press mentions, then generates a custom email referencing your actual AUM, the name of your auditor, and a recent client onboarding — in the sender's writing voice.
2. Real-time voice cloning
Deepfake voice calls have moved out of the research lab and into the field. Three minutes of public audio — a podcast appearance, a conference talk — is enough to clone a principal's voice convincingly. We've responded to two incidents this year where attackers used cloned voices during Zoom calls to authorize wire transfers.
3. Autonomous reconnaissance
AI agents can now perform the recon phase of an attack without human supervision. They crawl your public infrastructure, identify exposed services, cross-reference CVEs, and prepare exploit chains in minutes instead of days.
The most alarming trend isn't new attack vectors. It's the collapse of the reconnaissance-to-exploit timeline. What used to take a skilled human attacker a week now takes a scripted agent ninety minutes.
What actually helps
The good news: the fundamentals still work. Attackers are more efficient, but the gaps they exploit haven't fundamentally changed.
- Phishing-resistant MFA. Hardware security keys (FIDO2) or platform authenticators. SMS and TOTP are no longer sufficient for accounts that touch client data — and an examiner asking what your “reasonable design” means under 206(4)-7 will start with this.
- Out-of-band verification for any wire or standing-instruction change. If someone asks via email, voice, or video — verify through a different channel against a known number. Every time. Reg S-ID's red flags catch this if the workflow exists in writing.
- Continuous exposure management. Monthly vulnerability scans aren't enough anymore. You need continuous monitoring of your attack surface — and that monitoring needs to produce evidence your CCO can put in the file.
- Security awareness training that reflects 2026 reality. The old “check for spelling mistakes” advice is obsolete. Annual training is the floor; quarterly phishing simulations are where firms with mature programs are landing.
- Incident response tabletop exercises. Not annual. Quarterly. Include scenarios with AI-generated deepfakes — and walk the Reg S-P 30-day customer notice clock as part of the exercise.
Where this lands in the KNX stack
We've rolled out three new capabilities this year: AI-assisted anomaly detection in email flows, deepfake audio detection integrated into our unified communications monitoring, and an automated external attack-surface scanner that runs continuously. These sit inside the broader Cybersecurity Essentials service and feed evidence into Compliance & Training reports.
If you want to know where your firm stands, the Free Compliance Assessment is the cheapest way to find out.
Bottom line: The threat landscape has evolved faster than most security programs. The firms weathering this cycle well are the ones treating AI-aware attackers as the baseline assumption, not the edge case.