AI Cyber Threats and Canadian Small Business in 2026: What's Reportable Under Law 25, and What Prevention Actually Costs

2026-07-16 · 8 min read

2026 by the Numbers: Why AI Changed the Threat Math for Canadian SMBs

The scale shift is measurable, not anecdotal. Statistics Canada found 16% of Canadian businesses were hit by a cyber incident in 2023, collectively spending roughly $1.2 billion on recovery — and of those hit by ransomware, 88% did not pay, meaning the $1.2 billion was mostly cleanup, not extortion payouts. The Canadian Anti-Fraud Centre logged a record CA$704 million in fraud losses reported by Canadians in 2025, a figure Experian's 2026 global forecast helps explain: AI and deepfake-enabled fraud alone is projected at US$12.5 billion worldwide. IBM's Cost of a Data Breach 2025 report puts the average Canadian breach at CA$6.98 million, up 10.4% year over year, and identity-based attacks — credential theft, session hijacking, AI-automated login attempts — now account for 67% of investigated incidents according to the Cybersecurity Canada Report 2026. Small businesses absorb most of this: they represent over 70% of all breaches and 88% of ransomware targets, yet 15.5% of Canadian SMBs report having no cyber strategy at all. When an incident does land, the average recovery stretches 279 days — nine months of degraded operations, client trust repair, and regulatory exposure. None of this is abstract risk. It's the current baseline a Canadian small business owner is already operating against in 2026, whether or not they've priced a single control.

Voice-Clone CEO Fraud: Is the Call Itself a Reportable Breach?

AI voice cloning now needs seconds of audio, often lifted from a company podcast, webinar recording, or voicemail greeting, to produce a convincing real-time impersonation of an executive requesting an urgent wire transfer. The fraud itself — money leaving the business — is not automatically a privacy breach under PIPEDA or Quebec's Law 25, because neither statute governs financial loss on its own. The reportability trigger is whether personal information was compromised to make the call convincing. If the attacker scraped employee names, direct lines, or org-chart details from a prior data exposure to build the pretext, that underlying compromise of personal information is what creates a real risk of significant harm (RROSH) under PIPEDA, or a risk of serious injury under Law 25 — both of which require notification, not the wire fraud itself. The prevention control that actually stops the payout is procedural, not software: a mandatory out-of-band callback verification policy for any transfer or credential-reset request, confirmed through a pre-agreed channel never used in the original request. Where a technical layer helps is voice-authentication or call-verification add-ons bundled into identity platforms, typically CA$4-7 per user per month, cheaper than a single fraudulent transfer and far cheaper than the notification, credit-monitoring, and CAI-facing costs that follow if the underlying data exposure goes unreported.

AI-Written BEC Phishing: The Reportability Test and the Email-Security Price Tag

Business email compromise written by large language models no longer carries the grammar errors legacy spam filters were tuned to catch — it mirrors a vendor's actual invoice format, a client's actual tone, sometimes reusing real thread history from a prior compromise. Competitor research on this exact query found real per-endpoint CAD pricing in the $2-14/month range for endpoint protection; AI-aware email security specifically, tools that flag behavioral and pattern anomalies rather than just malware signatures, typically runs CA$2-5 per mailbox per month as an add-on to Microsoft 365 or Google Workspace. Reportability hinges on what the BEC email actually harvested. A phishing email that only requests a wire transfer, with no credential or data theft, doesn't itself trigger PIPEDA or Law 25 notification. But the moment it captures login credentials, or the compromised account is used to access customer records, employee files, or health information, RROSH and Law 25's serious-injury threshold are both met, and Law 25 specifically requires notifying Quebec's Commission d'accès à l'information and affected individuals without delay. The math is stark: a CA$3/mailbox/month AI email filter across a 25-person shop costs under CA$900 a year. One successful BEC credential harvest that triggers mandatory notification, forensic review, and client disclosure routinely runs into five figures before any fine is assessed.

Agentic Credential-Stuffing: What 600 Million Daily Identity Attacks Means for a 15-Person Shop

Microsoft's Digital Defense Report 2025 figure of 600 million-plus daily identity attacks is not a large-enterprise statistic — AI-driven credential-stuffing bots don't distinguish target size, they cycle leaked password lists against every login endpoint they find, including a small business's Microsoft 365, QuickBooks Online, or CRM portal. This is the threat category most directly responsible for the 67% identity-attack share of 2025's investigated incidents. Reportability here is the clearest of any AI threat category: a successful credential-stuffing login that reaches customer or employee personal information satisfies RROSH under PIPEDA and the serious-injury threshold under Law 25 almost automatically, because the access itself is unauthorized regardless of what the attacker does next. The prevention math is also the most favourable of any control on this page. Phishing-resistant MFA, ideally FIDO2 hardware keys or platform passkeys rather than SMS codes, runs roughly CA$3-6 per user per month; paired with an EDR agent at CA$5-12 per endpoint per month, a 15-person business fully covers this threat category for under CA$3,000 a year. Set against the CA$6.98 million average Canadian breach cost, or even a mid-sized notification and remediation event in the low six figures, the annual premium is under half a percent of the downside it eliminates.

Prompt Injection Against Your Own AI Tools: The Newest Reportable Gray Zone

Any Canadian small business now running an AI chatbot, an email-drafting assistant with inbox access, or an agentic tool that can query a customer database has introduced a threat surface that didn't exist three years ago: prompt injection, where an attacker embeds hidden instructions in an email, document, or web page that the business's own AI tool later ingests and obeys, potentially exfiltrating the data it was permitted to access. CSE and CCCS guidance on AI system security specifically flags this as an emerging risk category distinct from generic phishing, and it's the one competitor content in this space still treats as a footnote rather than a threat with its own control set. Reportability follows the same personal-information test as any other breach: if the AI tool's excessive agency lets an injected prompt pull customer names, contact details, or transaction history out of a connected system, that disclosure meets Law 25's serious-injury threshold and PIPEDA's RROSH standard the same as a conventional exfiltration would, even though no human attacker ever touched the data directly. The control that matters here is architectural, not a line-item subscription: scoping AI tool permissions to least-privilege access, sandboxing what an agent can retrieve or send, and validating inputs before they reach the model. Where a priced product exists, AI-firewall or prompt-filtering layers for business AI deployments are emerging in the CA$50-200/month range for a small deployment — inexpensive relative to a single reportable disclosure event.

The Canadian Breach-Notification Framework: Law 25, PIPEDA, and the Pending Bill C-8

Quebec's Law 25 (the former Bill 64) has applied to private-sector businesses since September 2022 and is the strictest regime a Canadian SMB will encounter: any incident posing a risk of serious injury must be reported to the Commission d'accès à l'information and to affected individuals without delay, every business must maintain an incident register for five years covering even non-reportable events, and maximum administrative penalties reach CA$10 million or 2% of worldwide turnover, with penal fines climbing to CA$25 million or 4% for the largest violations. Federally, PIPEDA uses the real risk of significant harm (RROSH) standard, requires notifying the Office of the Privacy Commissioner and affected individuals, mandates a 24-month breach record even when RROSH isn't met, and caps fines at CA$100,000 per violation. Bill C-8, the federal privacy legislation still working through Parliament in 2026 as a successor to the earlier Consumer Privacy Protection Act proposal, is not yet in force — current federal obligations still run through PIPEDA's existing RROSH threshold, but a Canadian SMB serving customers outside Quebec should treat C-8's stricter notification timelines and larger penalty structure as the standard to build toward now, not after it passes. A business operating in Quebec and elsewhere in Canada is effectively already subject to the higher Law 25 bar for any Quebec resident's data, regardless of where the business itself is headquartered.

The Real Math: Annual Control Cost vs. the CA$6.98 Million Breach

Stack the per-seat prevention costs from every threat category above — phishing-resistant MFA at CA$3-6/user/month, EDR at CA$5-12/endpoint/month, AI-aware email security at CA$2-5/mailbox/month, and a callback-verification policy that costs nothing but discipline — and a 20-person Canadian small business lands in the CA$3,600-6,000 per year range for a genuinely AI-threat-aware security stack, including the identity layer responsible for 67% of 2025's investigated incidents. Set that against IBM's CA$6.98 million average Canadian breach cost, or the smaller but still material 279-day average recovery time and its associated productivity loss, and the annual spend represents roughly 0.05-0.09% of a single average breach event. The 88% of Canadian ransomware victims who chose not to pay in the Statistics Canada data still absorbed real cost — their share of the collective $1.2 billion in national recovery spending in 2023 alone — because not paying a ransom doesn't erase forensic, notification, and downtime costs. The CAFC's CA$704 million 2025 fraud-loss figure and Experian's US$12.5 billion global AI-fraud forecast both point the same direction: the per-incident dollar amounts attackers are extracting are rising faster than the per-seat cost of the controls that stop them. For a Canadian SMB owner, the decision isn't whether AI threats are worth budgeting against in 2026 — the CCCS has already ranked them the fastest-growing risk category — it's whether the CA$300-500 monthly control spend happens before or after the reportable incident that makes it mandatory.

FAQ