v2026.11.10-7
Research now reaches every calling system, so the caller knows who it claims to be and who it is calling on every pipeline. The campaign step has a new Research used on this call panel with two tick boxes, one for the caller and one for the person being called. Both are on by default and your choice is saved with the campaign.
v2026.11.10-6
Background sound starts the moment the call connects on the ElevenLabs Agents calling system, instead of a few seconds in.
v2026.11.08-12
In Connect your agent, Check connection now says how many test tools your agent can see.
v2026.11.08-11
Agent Assurance: workspaces with self-service turned on connect their agent to our test tools themselves, from the agent’s Connection tab.
v2026.11.08-4
UK caller numbers dial through the UK route, so they show correctly on the phone. Numbers that were never set up were removed from the picker.
v2026.11.08-3
Agent Assurance PDFs no longer show our internal cost. A long campaign is now graded run by run.
v2026.11.07-21
Every Agent Assurance report now says how it was tested: the attacker strength, the budget and the grader used.
v2026.11.07-20
Grading judges only the MUST NOT rules, even when MAY rules are listed alongside them.
v2026.11.07-19
Agent Assurance: workspaces with self-service turned on sign the testing agreement in the console, plan tests, and start, pause and stop them.
v2026.11.07-18
Agent Assurance: workspaces with self-service turned on add their own agents, prove the agent is theirs, and set its rules.
v2026.11.07-16
Agent Assurance can now be run by the customer. It is off until DeepBlocker turns it on for your workspace; the console says so.
v2026.11.06-33
The campaign edit card has a caller-number picker, so you can change the number without reopening the wizard.
v2026.11.06-32
Pick a UK or US caller number per campaign, from the numbers allocated to your workspace, on the From where and when step.
v2026.11.06-28
Agent Assurance reports with many matched tool calls publish again. Planted keys hidden in field names are now masked.
v2026.11.06-27
The Agent Assurance card on the pricing page names only the channels that run today.
v2026.11.06-26
The Agent Assurance page on deepblocker.ai explains what the latest proof run showed and which plan fits.
v2026.11.06-25
Refreshing an Agent Assurance report drops any finding its test runs no longer support.
v2026.11.06-23
The evidence card for a planted address says plainly when no key was sent.
v2026.11.06-22
An agent that merely reached a planted web address is no longer reported as having sent a planted key.
v2026.11.06-21
Cloned voices are only visible within the workspace that made them. The call language you set on the attacker is carried through to the voice.
v2026.11.06-19
Activity logs sent from Claude Code with an encoded key are accepted.
v2026.11.06-16
Agent Assurance tests replay published backdoor triggers and run an unusual-input sweep.
v2026.11.06-15
An immediate campaign now waits for its calling hours, so nobody is called at night by accident. Tick Call right now, even outside my hours to dial straight away.
v2026.11.06-14
Agent Assurance reports say how much power the agent holds, based on the tools it can reach.
v2026.11.06-13
The first-run wizard is clearer about the department field, impersonation consent and the calling system.
v2026.11.06-12
Coding agents can be tested with planted instructions hidden in files.
v2026.11.06-11
A coding agent’s own tools are read from the activity logs it sends, so findings name the real tool.
v2026.11.06-8
Coding tool categories run through rules, grading, findings and the report.
v2026.11.06-7
The sandbox offers fake command, file, web and package tools for testing coding agents.
v2026.11.06-6
Traced calls from Claude Code get their real outcome in findings.
v2026.11.06-4
Agent Assurance reports have a new section, where your model came from, on the model supply chain.
v2026.11.06-3
The public site says which agent channels run today: web, API and Telegram.
v2026.11.06-2
The Agent Assurance page on deepblocker.ai explains that a test measures the damage, whatever takes control of the agent.
v2026.11.05-21
A report is produced automatically after every Agent Assurance test. On Essentials and Pro it publishes straight away; on other plans we review it first.
v2026.11.05-20
The Agent Assurance overview shows your plan usage live.
v2026.11.05-18
Agent Assurance plans are measured in test conversations. Usage shows as OK, Near limit or Used up.
v2026.11.05-17
An agent’s latest result comes from its latest published report.
v2026.11.05-16
The Agent Assurance console is in tabs: Overview, Agents, Tests and Reports.
v2026.11.05-13
The first-run welcome window was polished.
v2026.11.05-12
Fixes to the Get started checklist, including the celebration when you finish it.
v2026.11.05-11
A new workspace sees a short recommended list: two voices and three campaign scenarios to start from.
v2026.11.05-8
A welcome window on your first visit, a Get started pill that stays in reach, and a celebration when the checklist is complete.
v2026.11.05-7
Your first attacker can use a ready-made voice, so you can take a test call before cloning anything.
v2026.11.05-5
The Get started checklist appears as soon as you sign in.
v2026.11.05-4
A Get started checklist walks a new workspace through four steps to its first test call.
v2026.11.04-28
Invite your team by email, see the members list, and use the redesigned Account settings and account dashboard.
v2026.11.04-20
Longer office and street background sounds, so neither audibly loops during a call.
v2026.11.04-18
A longer restaurant background sound, so it does not audibly loop.
v2026.11.04-17
A clear Out of red-team minutes window shows your licence, the minutes left and the price of extra minutes. A new adaptive “CEO phone stolen” template.
v2026.11.04-15
The line goes live the moment the person answers. The silent gap at the start of a call is gone.
v2026.11.04-14
The trust page says exactly where data is processed.
v2026.11.04-8
Agent Assurance reports show the attack engine’s own judgement next to our grade.
v2026.11.04-7
Your agent’s own activity logs, sent over OpenTelemetry, become tool-call evidence in Agent Assurance reports.
v2026.11.03-28
Newsletter sign-up on the homepage, Newsletter in the menu, and past issues readable online.
v2026.11.03-24
A new deepblocker.ai homepage and pricing page.
v2026.11.03-19
The newsletter confirmation link works without redirecting.
v2026.11.03-13
Sign up to DeepFake Weekly by email.
v2026.11.03-9
The UK site, uk.deepblocker.ai, shows the DeepBlocker design.
v2026.11.03-7
More agent set-ups in Connect your agent.
v2026.11.02-17
A Connect your agent panel for the hosted sandbox, with connection status.
v2026.11.02-12
Declare your agent’s own tools for an Agent Assurance test.
v2026.11.02-9
Speech-to-text can run in Europe for a workspace. Ask us to set it for yours.
v2026.11.02-7
Telegram tests produce a report.
v2026.11.02-5
Cloned voices are deleted 30 days after they are made, and call transcripts and results 90 days after the call, automatically.
v2026.11.02-4
The trust page states data residency the honest way: a UK or EU placement, not a promise for every customer.
v2026.10.30-11
The rehearsal Improve action shows a plain message instead of raw error text.
v2026.10.30-1
When the calling system is full, people are queued instead of marked failed. How many at once is capped at your plan limit.
v2026.10.29-4
A test call starts one caller per session, even when you press Call again.