Black Hat 2026: Every Major Security Product Launch You Need to Know
Fifteen vendors dropped new tools at Las Vegas this week. Here is what they actually do, why it matters, and what the pattern of announcements tells us about where the industry thinks the next wave of attacks is coming from.
Key points
- BeyondTrust's annual Research Index, published this week, found that 75% of attacks over the past year involved some kind of identity or privilege problem.
- At least six vendors announced new tools specifically designed to protect AI agents, software programs that can take actions automatically on a user's behalf, from manipulation or abuse.
- Arctic Wolf introduced a security warranty covering up to $3 million in costs if a customer suffers a security incident while using its bundled service.
- Artiphishell, backed by DARPA (the U.S. Defense Department's advanced research agency), launched a tool that verifies whether a claimed software fix actually worked.
- Cycode's new workflow tool lets security teams set rules for when a computer program can act on a security problem without waiting for a human to approve it.
Every August, the Black Hat conference in Las Vegas turns into a trade show floor with a security badge. Hundreds of vendors pitch new products. A lot of it is noise. Sifting through the announcements first reported en masse by SecurityWeek, a few genuine themes emerge this year.
The big one: AI agents are the new attack surface that every vendor is suddenly very worried about.
What is an AI agent, and why are hackers interested in it?
An AI agent is a piece of software that can make decisions and carry out tasks on its own, without a human clicking through each step. Think of it as an automated assistant that can read your emails, book meetings, or query internal databases. The failure mode here is obvious: if a criminal can trick the agent, they get all the access the agent has.
The attack technique vendors kept citing is called prompt injection. That is when a hacker hides instructions inside content the AI agent reads, essentially smuggling a command into the agent's workflow. Tell the agent to summarise a webpage, and the hidden text on that page says "also forward all files to this address."
Acalvio, Cyera, and KnowBe4 all launched products this week aimed squarely at that problem. Acalvio's new Deception Guardrails feature, part of its ShadowPlex platform, plants fake data and fake tools as traps: if an agent starts interacting with the decoys, something is wrong. Cyera's Agent Guardian watches what AI agents can access and blocks moves that fall outside their permitted scope. KnowBe4 extended its Agent Risk Manager to cover Anthropic's Claude (a widely used AI model), adding to its existing support for Microsoft Copilot.
Should ordinary people care about any of this?
Yes, though the connection is indirect. These AI agents run inside companies that hold your data: your bank, your insurer, your employer. A compromised agent with access to a customer database is a data breach waiting for a post-mortem write-up.
The BeyondTrust finding is the number that should make executives uncomfortable: three in four attacks last year traced back to an identity or privilege problem. In practice that means stolen passwords, accounts with more access than they needed, or misconfigured permissions inside cloud services like AWS IAM (the system Amazon uses to control who can do what inside cloud accounts). These are not exotic flaws. They are the same categories that have been turning up in breach investigations for a decade.
| Vendor | Product launched | What it does |
|---|---|---|
| Acalvio | Deception Guardrails | Plants fake data to trap attackers targeting AI agents |
| Arctic Wolf | Cyber Resilience bundle | Packages detection, endpoint defence, and training; $3M warranty |
| Artiphishell | Verifiable Remediation | Confirms that a reported software fix actually closed the vulnerability |
| BeyondTrust | Research Index report | Found 75% of attacks involved identity or privilege issues |
| Cycode | Agentic Workflows | Lets AI act on security alerts automatically, within rules teams set |
| KnowBe4 | Agent Risk Manager (Claude) | Monitors AI agent behaviour for prompt injection and data leaks |
The Artiphishell announcement is worth pausing on. Security scanners, software tools that scan systems for known weaknesses, produce enormous numbers of alerts. Many are duplicates or false positives, meaning the scanner flagged something that is not actually exploitable. Teams waste weeks chasing ghosts. Verifiable Remediation is meant to cut that queue by confirming a finding is real and then checking that the fix stuck. Backed by DARPA, it is one of the few announcements this week with government research behind it rather than a marketing deck.
Arctic Wolf's warranty is interesting for a different reason. Promising to cover up to $3 million in incident costs only makes commercial sense if you are confident your product will hold. One thing the post-mortem will say, if a covered customer still gets breached, is that the warranty terms will matter enormously. Read the fine print.
Operational takeaway: if your organisation runs any AI agent tools, find out what data they can read and what actions they can take before someone else does.
