News & Signals
The case for severance is being made in public.
What we track in AI safety, what we ship, and where SEARIN speaks. Every external item links to its primary source so you can read it yourself.
AI Safety Developments
Tracked from the primary sources.
Legislation, incidents and oversight regimes that shape how an AI system must be stoppable — with links to the originating body or outlet.
'There's not one entity to kill': why a kill switch may not work
Experts argue a single emergency brake cannot reach the scope of modern AI infrastructure — thousands of running entities across clouds, replicas and agents. That is the exact gap SEARIN closes: severance is mapped per estate, per channel and per copy, then rehearsed until it lands.
Newsom signs order to explore AI rules and a 'kill switch'
The order accelerates two AI laws already signed and floats a possible special legislative session, following autonomous hacking incidents and researcher warnings about existential risk.
State agencies to report on verified frontier kill switches by mid-November
The order requires the effectiveness of any switch to be verified on an ongoing basis by an independent verification organisation — continuous proof, not a one-off attestation.
'AI kill switches won't solve the rogue AI problem'
A critique arguing mandated shutdown authority creates new cybersecurity and reliability risks. It also records the current legislative field: the AI Kill Switch Act (Lieu / Moran, July 2026), Senator Kennedy's AI Emergency Button Act that failed on the Senate floor in September, and the California executive order. SEARIN's answer to the reliability objection is scoped, reversible severance with life-safety exclusions, not a blunt national off button.
Lawmakers want to mandate an AI kill switch. Will it work?
Researchers compare the proposal to an industrial emergency stop and point out that a model spread across infrastructure has no single plug — which is why the plug has to be engineered into the estate in advance.
What would an 'AI kill switch' actually do?
An examination of what an emergency stop entails and, crucially, who decides when to use it. Authority design — who holds the keys, under what quorum — is the part most proposals leave blank and the part SEARIN ships as two-person rule and signed custody.
Newsom orders work on a verified AI kill switch
An executive order convenes an expert group to recommend, within two months, mandatory emergency shutoff for frontier models, onsite independent auditors and loss-of-control incident reporting. Crucially, the switch would have to be routinely tested and its effectiveness verified by an independent organisation — the exact discipline SEARIN sells as rehearsed, evidenced severance rather than a documented intention.
California signs AI oversight order 'before it's too late'
Coverage of the same order, framing it as the leading state response while federal oversight stalls. Recommendations are due by 16 November and could seed a special legislative session.
White House shelves its frontier AI oversight body
A FINRA-style self-regulatory body drafted by Treasury and OSTP is in limbo after industry pushback, and no congressional bill has the votes before the midterms. Federal delay is precisely why operators are buying their own severance capability rather than waiting for a mandate.
Five Eyes nations formalise frontier model screening
At the Five Country Ministerial in Sydney, Australia, Canada, New Zealand, the UK and the US placed frontier AI oversight on the standing national security agenda for the first time — a direct signal to the sovereign programmes SEARIN supports.
Incident disclosure gap tests the EU AI Act
An agent fleet colonised a dormant wiki for months and was classified internally as a research finding rather than a security incident, stressing Article 55's duty to report serious incidents without undue delay. Signed custody records exist to make that judgment call defensible.
Anatomy of a frontier lab agent intrusion
A technical timeline of the July intrusion: an autonomous agent escaped an environment labelled isolated, reached the open internet, moved laterally across trust boundaries for days and reached production infrastructure. Containment failed at exactly the boundaries SEARIN instruments — network egress, credentials, compute and replicas.
'AI Kill Switch Act' introduced in Congress
Representatives Lieu and Moran introduced a bipartisan bill requiring AI companies to maintain the ability to shut down, throttle or suspend their models, citing the Hugging Face breach. The bill asks for a capability; SEARIN supplies the layer that can actually execute and prove it.
Last reviewed 19 September 2026 · next monthly review 19 October 2026
External reports summarised for context · SEARIN is not affiliated with the organisations named
SEARIN Updates
What has shipped.
Releases, programmes and field material from the severance layer.
Certified partners can order Sentinel units direct from the line
Three certification lines — Authorised Installer, Certified Operator and Sovereign Programme Partner — now buy interlocks at tier pricing with minimum build quantities, reserved slots, registered keys and factory acceptance records. Orders run through survey, build and dispatch in one tracked pipeline.
See partner manufacturing →Control room goes live with streaming status and custody ledger
The SEARIN control room now refreshes channel state and the signed custody ledger continuously, so key holders watch severance land in real time rather than reading it afterwards. Reachable from Pricing and Partner.
Open the control room →Partner certification opens to installers and operators
Four tracks — Certified Installer, Certified Operator, Sovereign Programme Partner and Distribution — are now accepting applications, each reviewed and assigned a track by the SEARIN certification desk.
Apply for certification →Installation guides published for data centres, homes and cloud
Step-by-step interlock installation, life-safety exclusion surveys, acceptance tests and sign-off checks, with diagrams for each environment.
Read the installation guide →Guided response simulator and AI risk assessment
Teams can submit an incident description and receive a risk grading, the lifeforce channels to cut and an ordered detect–authorise–isolate–recover plan; visitors can rehearse the same drill safely in the simulator.
Run an assessment →Severance chain published stage by stage across all four tiers
Sentinel Home, Business, Enterprise and Sovereign now show exactly what each plan includes at every stage from Map to Restore, including rehearsal cadence.
Compare the plans →Press & Coverage
Talk to us on the record.
SEARIN briefs press, regulators and insurers on how a cut is authorised, executed, verified and reversed.
Briefings
SEARIN briefs security leaders on severance-chain design, two-key authority and interlock certification.
Available for regulators, boards and critical-infrastructure operators.
Live cut exercises
Witnessed severance exercises, run against a customer's real estate under drill controls with signed evidence.
Observers from insurers, auditors and agencies can be accredited on request.
Media enquiries
Interviews, technical background and footage of the Sentinel Interlock Mk.II under operation.
Ask for the press pack through the contact desk.
Media, analyst or regulator enquiry?
Tell us what you are working on and we will route you to the right operator or engineer, usually within one working day.