Skip to content

Topic

Confidential Computing

Security approach using hardware trusted execution environments to protect data during processing, isolating it from the host, hypervisor, or cloud provider.

Current stories

buildOne report1 publisher

Anthropic licenses Opaque's confidential-AI technology and hires its CEO and CTO, The Information reports

Anthropic is licensing Opaque Systems' technology and hiring some staff, including its CEO and CTO, The Information reported. Neither company has detailed the terms, including what they mean for Opaque's open-source tooling.

Publishers:thenewstack.io

Reality

Evidence50
Adoption
Not scored
Hype gap+10
Incentives
Not scored
Confidence55
buildOne report1 publisher

Intel ends SGX at Xeon Diamond Rapids and puts its confidential computing work on TDX

Intel said Xeon Diamond Rapids will be the last generation to support SGX, Tom's Hardware reported. Existing SGX platforms stay supported into the early 2030s, so an enclave team's nearer deadline is its first Xeon refresh after Diamond Rapids.

Publishers:tomshardware.com

Reality

Evidence55
Adoption60
Hype gap+15
Incentives60
Confidence65
productOne report1 publisher

Iterate.ai's own test shows Lifeboat doubling agent sessions on a single GPU

Iterate.ai launched Lifeboat, an inference engine it says fits two to six times more AI agent sessions on each GPU. Its own published test shows a doubling on one Nvidia card, still enough to make regulated firms measure the cards they own before buying more.

Publishers:siliconangle.com

Reality

Evidence35
Adoption15
Hype gap+45
Incentives75
Confidence40
productOne report1 publisher

Meta says its Muse agent may still remember what it learned from deleted chats

Meta says its Muse agent "may still remember information it learned from what you deleted," and PCWorld reports that Muse data trains Meta's AI by default. Anyone connecting an inbox to it is making a choice that deleting a chat will not fully reverse.

Publishers:pcworld.com

Reality

Evidence45
Adoption
Not scored
Hype gap
Not scored
Incentives
Not scored
Confidence40
buildOne report1 publisher

Google moves federated learning onto attested servers so auditors can check the privacy noise

Google Research now trains federated models inside attested TEEs and logs each access policy to the public Rekor log, with Gboard already switched over. The server-side noise step that users once had to take on trust becomes code an outside auditor can check against open source.

Publishers:research.google

Reality

Evidence45
Adoption35
Hype gap+20
Incentives60
Confidence50
scienceOne report1 publisher

Anthropic's reset provable-inference deadline lapses without a public update

Anthropic's Sept 30 deadline for Phase 1 of its provable-inference project, already pushed back 138 days, passed with no public update. The lapse shows how little of a dated pledge in its Responsible Scaling Policy anyone outside the company can check.

Publishers:forkast.news

Reality

Evidence32
Adoption
Not scored
Hype gap+20
Incentives
Not scored
Confidence35
productOne report1 publisher

Meta's first camera-less Ray-Bans arrive as venues ban smart glasses

Meta unveiled Ray-Ban Audio, its first smart glasses without a camera, at Connect alongside a third generation of its camera Ray-Bans. It is the one Connect privacy change a venue's device rule can act on, since Private Processing has not reached the glasses yet.

Publishers:gizmodo.com

Reality

Evidence40
Adoption
Not scored
Hype gap+25
Incentives55
Confidence40
securityConfirmed4 publishers

TRACE goes to the Linux Foundation, and the audit question becomes: show me the artifact

The Linux Foundation is taking over an open spec for hardware-signed agent runtime evidence. The producing side now has code and a spec; nobody has yet named who accepts the record.

Perspective Coverage

4 publishers
Builder
Builder 39%
Operator
Operator 46%
Investor
Investor 15%

Reality

Evidence55
Adoption20
Hype gap+30
Incentives70
Confidence60
leadershipConfirmed4 publishers

Meta ships Muse with email and payment access; the user-held encryption key is not out yet

The agent launched in the US on Tuesday with subscriptions at $20 and $100 a month, and employees testing it during launch week were still logging guardrail bypasses, logouts the system forced on its own, and monitoring that occasionally shut off without explanation.

Publishers:about.fb.com implicator.ai newsable.asianetnews.com skift.com

Perspective Coverage

4 publishers
Builder
Builder 35%
Operator
Operator 45%
Investor
Investor 20%

Reality

Evidence55
Adoption
Not scored
Hype gap+40
Incentives65
Confidence60
securityOne report1 publisher

Google plans to store Private AI Compute context in per-user databases keyed to the user's device

Private AI Compute has erased all context at the end of each task. Google plans to keep per-user context in encrypted cloud storage instead, with the decryption keys held only on the user's own devices.

Reality

Evidence44
Adoption
Not scored
Hype gap+22
Incentives66
Confidence52
buildOne report1 publisher

Apple relocates image provenance from the edit chain to the iPhone sensor

Reference mode on the iPhone 18 Pro signs pixels as they leave the sensor and finishes the picture inside Private Cloud Compute, so anyone building a verification pipeline now has an Apple trust root to handle alongside C2PA.

Publishers:infoq.com

Reality

Evidence56
Adoption24
Hype gap+28
Incentives72
Confidence57
buildOne report1 publisher

Meta puts glasses AI inside confidential VMs the hypervisor can only read as ciphertext

Meta says a stateful, deeply personal assistant cannot fit in a pair of glasses, so inference runs in confidential VMs in its data centres, and the client sends nothing unless two attestation checks pass.

Reality

Evidence34
Adoption25
Hype gap+24
Incentives86
Confidence51
productOne report1 publisher

Meta's glasses privacy platform goes to auditors before it goes to wearers

Private Processing, the siloed cloud scheme Meta built for WhatsApp last year, is coming to Ray-Ban Meta glasses so the assistant can work on data Meta says it cannot read. Meta's engineering director told WIRED he does not have a timeline.

Publishers:wired.com

Reality

Evidence46
Adoption24
Hype gap+42
Incentives71
Confidence56
buildOne report1 publisher

Confidential inference on Blackwell retains 96-98 percent of throughput with CC-aware adaptations, NVIDIA reports

NVIDIA measured TensorRT LLM holding 96.1 to 98.2 percent of its non-confidential output throughput on Blackwell, and it got there by unpinning host memory on the affected paths, moving decode readback off the scheduler thread, and timing kernel tactics with the GPU's global timer instead of CUDA events.

Reality

Evidence58
Adoption22
Hype gap+10
Incentives80
Confidence55
productOne report1 publisher

Vast Data hands banks and model vendors separate keys to the same GPU enclave

DataEnclave uses Nvidia confidential computing so a model developer's base weights and a customer's fine-tuned weights can run in one enclave under separate key control. Vast says it charges nothing extra for it.

Publishers:siliconangle.com

Reality

Evidence38
Adoption15
Hype gap+30
Incentives80
Confidence45
securityOne report1 publisher

Apple signs iPhone photos inside the camera hardware before software touches the pixels

Reference Image on the iPhone 18 Pro puts one signature on the sensor's raw output and a second on the developed file. The capture time it proves is only good to about 15 minutes.

Publishers:cyberinsider.com

Reality

Evidence48
Adoption8
Hype gap+20
Incentives62
Confidence55
securityOne report1 publisher

DDRop silently drops DDR5 writes so Intel TDX reads stale memory as current

Confidential computing never promised memory freshness. DDRop turns that omission into arbitrary access inside a TDX guest, using a board that costs under $200 and two prerequisites a cloud provider already holds.

Reality

Evidence62
Adoption22
Hype gap+20
Incentives45
Confidence58
buildOne report1 publisher

Zuckerberg expects Muse to collect a small cut from the businesses it transacts with

Meta is seeding its new personal agent with a large free-compute allowance and expects to be paid later by the businesses whose sales the agent completes. The fee rate, merchant terms and rollout timing remain undefined.

Publishers:runtimewire.com

Reality

Evidence55
Adoption20
Hype gap+20
Incentives72
Confidence55

Earlier coverage

  1. Apple's Audio Intelligence puts the consent switch on one wrist

    Product · September 10, 2026 · Confirmed21 publishers