2026, the perfect time to reflect on how far technology has come and what lies ahead. Without a doubt, Artificial intelligence has gone from a niche to an omnipresent force, reshaping how we work, build, and defend. While organisations have speed-ran the adoption of AI and machine learning, cybercriminals have been just as fast to exploit them, and AI now powers business decisions, customer interactions, and – predictably – cyberattacks.

In this post, I’ll reflect on what changed over the past two years, and look ahead to how we can prepare, secure our environments, and make sense of the AI-driven world we now operate in.

The last two years saw an explosion in AI adoption. Businesses integrated AI models into their day-to-day operations, users and employees integrated it into their daily lives, and threat actors leveraged AI to improve their attacks. And the attackers have quite a lot to show for it:

Software supply chain breaches escalated with the rise of AI. The number of malicious open-source packages has increased 

, with a growing focus on the poisoning of AI libraries and models. JFrog’s 2025 

 documented over 1 million new models added to Hugging Face in 2024 alone, with a 6.5x increase in malicious models uploaded by attackers.

 malware, able to rewrite itself and adapt to the target environment. Security researchers noted that modern malware can now automatically evade sandbox environments and blend into normal operations semantically, 

Attackers also used AI to improve their social engineering efforts. For example, AI-driven voice cloning made phone scams far more convincing, with voice phishing attacks having increased 

 by late 2024, according to CrowdStrike data. Auto-generated spear-phishing emails, deepfake audio and video, and choreographed AI scam campaigns increased in frequency and by mid-2025, criminals were breaching networks and starting to move laterally in just 

Uncontrolled AI Adoption and Shadow Systems

As organisations rushed to deploy AI (outpacing governance and security controls), a 2025 survey found that 

 of security teams could not identify where LLMs were running, and 

 of organisations allowed employees to deploy AI technologies without approvals. This lack of oversight led to widespread “shadow AI”, exposing sensitive data and introducing unvetted models and technologies into critical systems. The impact of this was evident in an EY report showing that 

 have already suffered AI-related security issues, with only 14% of CEOs believing that AI adequately protects sensitive data.

Early Response Regulation and AI-Aware Security

By late 2025, we’d started to see a few adjustments to this new reality. Forward-leaning organisations adopted AI-aware security tooling, including implementing scanners for malicious models, and AI usage control platforms (like CultureAI) to manage secure adoption across the workforce. These tools provided visibility into shadow AI use, enabled sanitisation of prompts, and supported policy enforcement around AI use and data handling.

Meanwhile, industry and regulatory bodies accelerated their efforts to formalise AI risk management:

By the end of 2025, AI had clearly redefined the threat landscape, and the organisations best positioned to adapt were those that treated AI governance, usage control, and secure-by-default tooling as essential, not optional overhead.

We now understand the scale of the challenge, and 2026 will be about operationalising that understanding, putting frameworks, tooling, and governance into practice before the next wave of attacks arrives. For security leaders, preparation is no longer about distant future proofing. It is about acting now to protect against AI-driven threats that already exist.

So, based on the lessons of ~2024 -2025, what are a few clear priorities that have emerged for 2026 and beyond?

Attackers are increasingly using AI for polymorphic malware, model poisoning, and automated attacks. Defenders should expect continuously adapting threats and shift away from static signatures towards behaviour-based detection. A large-scale AI model supply chain breach is plausible in 2026. Organisations with model inventories and integrity checks will be best placed to respond quickly.

AI regulation will become an enforceable reality. The EU AI Act is in force, and the Cyber Resilience Act will require SBOMs for software, including AI components. In parallel, organisations will increasingly align their internal AI governance with emerging standards such as ISO/IEC 42001. As a result, AI risk management will become a standard part of enterprise risk frameworks, driven by regulators, customers, and insurers alike.

AI-powered security tools will see wider adoption to cope with the scale and speed of AI-driven threats and adoption. This will include malware analysis, model scanning, and AI usage control solutions. This will help organisations enable AI use while preventing data leaks, enforcing policy, and providing guardrails around everyday AI interactions. Human oversight will remain critical. As defensive AI can be manipulated or poisoned, organisations must adopt a human-in-the-loop model with skilled analysts in control, continuously validating AI outputs.

Beyond regulation and internal governance, the broader AI security ecosystem will continue to mature. NIST is expected to finalise its AI Cybersecurity Profile, while industry groups expand shared tooling for model transparency, AI asset/dependency inventory generation, and vulnerability scanning. These developments will lower the barrier to adoption and create greater consistency across organisations. AI-specific requirements are likely to be embedded into security policies, assurance processes, and training as part of normal security operations.

Taken together, these trends point to a clear direction of travel. AI-driven threats will intensify, regulation will harden, and expectations around governance and control will rise regardless of organisational readiness. The question for security leaders is no longer whether these changes will matter, but how quickly they can be operationalised.

For organisations that have not yet put these capabilities in place, the following steps outline practical ways to begin securing AI-driven environments.

Taken together, these steps apply established security principles to AI systems. They reduce risk while enabling confident adoption of AI. Organisations that build trust and resilience into their AI environments will be better positioned both defensively and competitively in the years ahead

AI has become embedded in the very fabric of modern society, shaping how work is done, how people think, and how attacks unfold. The rapid adoption of AI over the past few years revealed huge gains, but often without the visibility, governance, and controls required to manage the newly introduced risks. Attackers adapted quickly, AI supply chains emerged as viable targets, and many organisations found their users operating AI systems that were unaccounted for and unvetted.

The challenge now is no longer about anticipating AI’s impact but enabling the secure operation of an AI-dominated environment. Frameworks are maturing, and expectations from regulators focus on the need for demonstrable control and accountability.

Organisations that are succeeding in 2026 are those enabling AI safely rather than slowing adoption, investing in visibility, usage control, and secure-by-default practices across the AI lifecycle. Human oversight remains essential, as effective defence still depends on informed judgement alongside automation.

AI is no longer an emerging technology to be planned for. It is a present reality that must be governed, secured, and operated with intent.

2026, the perfect time to reflect on how far technology has come and what lies ahead. Without a doubt, Artificial intelligence has gone from a niche to an omnipresent force, reshaping how we work, build, and defend. While organisations have speed-ran the adoption of AI and machine learning, cybercriminals have been just as fast to exploit them, and AI now powers business decisions, customer interactions, and – predictably – cyberattacks.
In this post, I’ll reflect on what changed over the past two years, and look ahead to how we can prepare, secure our environments, and make sense of the AI-driven world we now operate in.


AI 2026: A Look Ahead

Recommended for you

Scattered Spider and DragonForce: A Case Study in Human-Centric Cyber Threats

AI Browsers: A Security NightmareFlipping the Board on Decades of Security Progress

Case Study: How a Digital Bank Reduced Shadow AI Risk by 80% — Without Blocking Innovation