E74 - Welcome to 2026

What You Need to Know  

Welcome to 2026—a new chapter for FIR Risk Tuesday. This year brings more content, more detailed risk intelligence to help you navigate our rapidly changing world.

We start with Google’s Cybersecurity Forecast 2026 report, analyzed through the FIR Risk Platform to share insights that will help you plan for the year ahead.

The headline: AI isn’t just accelerating cybersecurity—it is cybersecurity. Attackers are scaling deception at machine-speed, targeting AI systems directly, and exploiting cloud and critical infrastructure with unprecedented precision. Defenders must now secure models, agents, people, and pipelines with equal rigor.

AI Is Now an Attack Surface, Not Just a Tool  

One of the biggest shifts happening under the hood is how attackers are beginning to treat AI systems—not as obstacles to bypass, but as systems to exploit in their own right.

Prompt injection, once an academic curiosity, is now a practical threat: malicious inputs crafted to coax chatbots into releasing sensitive data or running harmful logic. Model poisoning is entering the mainstream—especially as organizations integrate more third-party datasets into training pipelines. A poisoned training set today becomes a backdoored decision tomorrow.

And then there are shadow agents—AI processes operating inside trusted environments, quietly automating things they shouldn’t be automating. Exfiltrating files. Rewriting configs. Making decisions no one approved. This is where we begin to see AI as both “employee” and “insider threat.”

INTEL [GLOBAL THREAT]: AI-generated deepfake audio will be used in 40% of BEC attacks by 2026, with average losses surpassing $500K per incident.

When the voice on the phone sounds exactly like your CFO, the burden of proof shifts dramatically.

Deepfakes, Voice Cloning, and the Reinvention of Social Engineering  

The social-engineering landscape is changing faster than any traditional control can keep up with. Generative AI now scrapes public profiles—LinkedIn, GitHub, business bios—and uses that data to craft spear-phishing messages indistinguishable from genuine communication.

Calls and video messages are no longer “odd” or “off” when they’re faked. They’re convincing. They’re timely. They’re tailored.

And with AI capable of performing reconnaissance at scale, attackers don’t need to break into organizations—they can simply talk their way in.

Autonomous AI Agents: When Good Logic Goes Bad  

Organizations are beginning to adopt AI-driven workflows powered by chain-of-thought engines, autonomous agents, and advanced orchestration. This brings efficiency, sure. But it also introduces new classes of risk:

• AI that misinterprets prompts and takes unintended actions
• Third-party AI APIs that leak or retain sensitive data
• Autonomous logic loops that escalate privileges or trigger unintended changes
• Vendors whose own AI infrastructure becomes your attack vector

These aren’t hypothetical; they’re happening.

INTEL [GLOBAL RECOMMENDATION]: Organizations should implement NIST AI RMF controls to harden AI models against adversarial threats.

The AI supply chain isn’t just emerging—it’s already sprawling.

Where MITRE ATT&CK Fits Into the Story  

While MITRE continues adapting to AI-era threats, many familiar categories still apply—just at machine scale.

The tactics remain recognizable—the speed and precision do not.

Cybercrime’s Broader Shape in 2026  

While AI grabs headlines, traditional cybercrime isn’t slowing down—it’s evolving.

Ransomware 2.0 emphasizes double extortion and rapid compromise. The on-chain cybercrime economy grows more sophisticated, with attackers leaning harder on privacy coins and obscure decentralized exchanges. Cloud jacking—taking advantage of misconfigured multi-cloud environments—continues to gain momentum. And virtualization layers, once assumed safe, become compelling targets as one exploit can collapse an entire cluster.

INTEL [INDUSTRY PATTERN]: Financial institutions are 3x more likely to face AI-driven ransomware attacks than other sectors.

Meanwhile, ICS/OT environments face their own high-stakes risks: from wipers to critical infrastructure sabotage.

So What Should Organizations Actually Do?  

This part doesn’t need theatrics—only clarity.

1. Secure your AI systems. Harden training pipelines. Validate inputs. Restrict agent permissions. Monitor for anomalies that don’t look “malicious” in a traditional sense but absolutely are in an AI context.

2. Reinforce the human layer. Train employees to pause when voices sound familiar but the requests feel off. Simulate deepfake-enabled attacks. Make verification normal, not suspicious.

3. Adopt threat-informed defense. Use intelligence. Map your controls to ATT&CK. Prioritize T1059.006 (AI amplification) and T1566 (phishing), not because they’re new, but because they’re becoming dramatically more effective.

4. Align with governance frameworks. The NIST AI RMF isn’t a compliance document—it’s a safety net.

And finally: Assume AI will be involved in every incident—on both sides. Because in 2026, it will be.