Heading into the final quarter of 2025, the security story is speed. Attackers are using AI to automate more of their work, exploiting zero-days and misconfigurations faster than traditional patch and review cycles can respond. At the same time, trust itself has become an attack surface: deepfake-driven social engineering targets human judgment, and data poisoning targets the machine learning models businesses are starting to rely on.
Defenders are responding with behavioral analytics, threat intelligence that connects endpoint, cloud and operational technology (OT) signals, AI-assisted security operations, zero-trust maturity programs and early post-quantum cryptography pilots. Below are our five predictions for Q4 2025 and into 2026, what each means for a mid-sized business, and the actions that matter most.
Prediction 1: AI-assisted attacks compress the time to exploit
Through 2025, threat intelligence teams at Google, Anthropic and others documented attackers using AI models for reconnaissance, vulnerability research, phishing content and malware troubleshooting. The consistent finding was not that AI invented new attacks, but that it made existing ones faster and cheaper. We expect that trend to accelerate in Q4, with more activity aimed at internet-facing edge devices such as firewalls and VPN gateways, where a single unpatched flaw offers direct access.
What it means: a monthly patch cycle for exposed systems is no longer adequate. Critical fixes for anything reachable from the internet should be measured in days, prioritized against CISA’s Known Exploited Vulnerabilities catalog. Our piece on AI-powered threats walks through each stage of the attack chain in more depth.
Prediction 2: trust becomes the primary target
Deepfake audio and video have moved from novelty to fraud tool. The widely reported 2024 case at engineering firm Arup, where an employee transferred roughly $25 million after a video call with deepfaked colleagues, remains the clearest example. In 2025 the FBI warned that criminals were using AI-generated voice messages to impersonate senior US officials. Help desks, finance teams and executive assistants are the most exposed roles.
The parallel threat is algorithmic trust. Data poisoning, where an attacker plants manipulated content in data that a model learns from or retrieves, can cause AI systems to behave incorrectly in ways that are hard to detect. As businesses connect AI assistants to internal documents, email and the open web, the quality and provenance of that data becomes a security question, not only a data-quality one.
Prediction 3: security operations become AI-assisted by default
Security teams cannot hire their way out of alert volume. We expect most security operations platforms and managed detection providers to ship AI features for alert triage, investigation summaries and suggested response steps by the end of 2025. The value is real: faster triage and fewer missed signals. The risk is over-trust. A model that confidently summarizes an incident can also confidently miss one.
The right model is human-in-the-loop. Automate high-confidence, reversible actions such as isolating a laptop or disabling a suspicious session, and keep people in charge of judgment calls, threat modeling and anything that affects customers or production systems. Continuous validation, such as regular attack simulation and control testing, tells you whether the automation is actually working.
Predictions 4 and 5: zero trust matures and post-quantum planning begins
Zero trust is shifting from slogan to measured program. Frameworks such as CISA’s Zero Trust Maturity Model, whose version 2.0 was released in 2023, give organizations a way to score themselves across identity, devices, networks, applications and data, and to plan improvements in stages rather than as a single project.
Post-quantum cryptography is the longer-horizon item. NIST published its first three post-quantum standards (FIPS 203, 204 and 205) in August 2024, and its draft transition guidance, NIST IR 8547, proposes deprecating today’s widely used public-key algorithms such as RSA and elliptic-curve cryptography by 2030 and disallowing them by 2035. The concern is “harvest now, decrypt later”: data stolen today could be decrypted once a capable quantum computer exists. Most businesses do not need to migrate yet, but they do need to know where their cryptography lives.
What to do before the year ends
For a mid-sized organization, these predictions translate into a short, practical list for the quarter:
- Inventory internet-facing assets and set a days-not-weeks patch target for critical fixes on them.
- Add out-of-band verification for payment changes, wire requests and help-desk credential resets.
- Roll out phishing-resistant MFA (passkeys or FIDO2 keys) for administrators and finance staff first.
- Document which data your AI tools can read, who can change that data, and how outputs are reviewed.
- Score yourself against a zero-trust maturity model and pick the two weakest areas for 2026 budgeting.
- Start a cryptographic inventory and ask key vendors for their post-quantum roadmaps.
The trade-off is attention. None of these items is glamorous, and each competes with year-end projects. But each closes a gap that attackers are actively using now. The organizations that lead on resilience will be the ones that pair AI-augmented defense with continuous validation and human judgment, and that move from reacting to incidents toward anticipating them.
Update (September 2026): Several of these predictions arrived sooner than expected. Days after this post, Anthropic, the UK AI Security Institute and the Alan Turing Institute published research showing that as few as 250 malicious documents could implant a backdoor in language models across a range of sizes, making data poisoning a practical concern rather than a theoretical one. In November 2025 Anthropic reported disrupting a largely AI-orchestrated espionage campaign it attributed to a Chinese state-sponsored group. The action list above still applies.
Frequently asked questions
Is post-quantum cryptography urgent for a small business?
Migration is not urgent for most, but planning is. Knowing which systems use which encryption, and which vendors will deliver post-quantum support, makes the eventual transition far cheaper.
Should we let AI take automatic action in our security tools?
Yes, for high-confidence, reversible actions such as isolating a device. Keep human approval for anything that is hard to undo or affects customers.
What is the single most effective defense against deepfake fraud?
A verification procedure that does not depend on recognizing a face or voice: call back on a number already on file before moving money or resetting access.
Plan your next quarter
Delana Technologies helps businesses turn security predictions into a prioritized plan, from patching and identity to zero-trust maturity and post-quantum readiness. Learn about our cybersecurity and compliance services, call 239.414.5126 or contact us.
Sources: NIST FIPS 203, 204 and 205 (August 2024); NIST IR 8547 initial public draft, “Transition to Post-Quantum Cryptography Standards” (November 2024); CISA Zero Trust Maturity Model v2.0 (2023); Google Threat Intelligence Group, “Adversarial Misuse of Generative AI” (January 2025); FBI public service announcement on impersonation of senior US officials (May 2025); CNN reporting on the Arup deepfake fraud (2024); Anthropic, “A small number of samples can poison LLMs of any size” (October 2025); Anthropic, “Disrupting the first reported AI-orchestrated cyber espionage campaign” (November 2025).