Oblivious

• OBLV Deploy runs on AWS Nitro Enclaves with Azure in preview, so the vendors that own the enclave hardware could add the Kubernetes attestation and key-release workflow to it themselves.
• No private-sector customer appears by name in the reviewed sources, so a private-sector buyer cannot check a reference inside its own sector.
• The reviewed sources carry no revenue figure, customer count or funding round after April 2023, which is thin ground for a multi-year commitment.
• Oblivious told OpenDP that public differential privacy libraries are the backbone of its disclosure controls, so its advantage is the budget accounting and workflow around them rather than the privacy mechanisms.
• Oblivious published its differential privacy auditing tool as free software, so the auditing method is available to competitors and is not an exclusive artifact.
• The ISO 27001 and SOC 2 marks on the homepage carry no public trust portal or report behind them, so a buyer's security review starts from a request rather than from published evidence.

Oblivious sells against a gap its own pages describe plainly. Encryption protects data at rest and in transit, and the homepage argues that those controls fall away the moment data is queried, analysed or fed to a model. The homepage names financial institutions moving card and transaction data and engineering and data teams migrating high-risk workloads to the cloud, and the Antigranular pages name the administrators, data owners and data scientists who work through the AGENT console.

The independent record frames the same problem from the public-sector side. Silicon Republic reported in January 2022 that the United Nations was launching a virtual lab to unlock international data sharing, with national statistics offices from the United States, the Netherlands, Italy and the United Kingdom taking part, because those bodies hold data they may not share directly.

The pain is sized only in the broadest terms. Silicon Republic cited a 2019 McKinsey estimate that connecting data across institutional and geographic boundaries could create roughly $3trn in annual economic value, and no cited source narrows that to what blocked data access costs the buyers Oblivious sells to. [s1, s3, s11, s5]

OBLV Deploy is the confidential computing half. It runs existing applications inside hardware secure enclaves without code changes, and it releases decryption keys only after cryptographic attestation confirms which software and configuration are running. The product runs on Kubernetes with standard tooling, and the vendor lists AWS availability with Azure in preview.

AGENT is the differential privacy half. It is a middleware layer between enterprise data sources and the people querying them, enforces differential privacy on results, and tracks a privacy budget for each user and team against every sensitive source. Data scientists reach it through Jupyter notebooks and SQL clients, and the vendor lists pre-built connectors to Snowflake and Salesforce.

The documentation goes past marketing depth. The OBLV Deploy docs cover enclave lifecycle, AWS Nitro Enclaves, manifest and allowlist configuration, key management against an external KMS, certificate handling, autoscaling and an attestation proxy, which is the level of detail an engineer would need to deploy it. [s3, s4, s5, s6]

Oblivious competes in a crowded confidential computing field on ergonomics rather than on the underlying isolation. The hardware is AWS Nitro Enclaves and equivalent cloud offerings, which any rival can also build on, so the differentiator it argues for is that applications need no modification and that standard Kubernetes tooling keeps working.

The differential privacy side depends on public foundations too. Oblivious told Harvard's OpenDP that the OpenDP library and SmartNoise have been the backbone of its disclosure controls, and its community platform integrates OpenDP, SmartNoise and DiffPrivLib. A competitor can adopt the same libraries.

CB Insights lists Mithril Security as a top competitor and places Fortanix in the comparison set it generates for Oblivious. Neither the vendor's pages nor the cited press name a rival, so the competitive picture in the reviewed record comes from a third-party database rather than from head-to-head evidence. [s3, s12, s5, s17]

Publicly named traction is institutional, and the commercial work the sources describe is identified only by sector. Silicon Republic reported that it joined the UN PET Lab programme alongside OpenMined, Harvard's OpenDP published a use-case interview describing joint work on a UN trade data pilot and on a project with an unnamed major US telecom company, and TikTok's developer blog records the PET Arena capture-the-flag competition the two launched in December 2025.

Tech.eu reported that Oblivious counts telecom giants, the United Nations, official statistics offices, AI startups and privacy-technology communities among its clients. No cited source names a private-sector customer, and the homepage says Oblivious supports teams across these organisations without naming any of them on the page.

Community building is the visible demand-generation motion. Oblivious runs the annual Eyes-Off Data Summit and the Antigranular competition platform, and it told OpenDP that a 2022 UN hackathon drew over 300 participants from 33 countries. [s11, s12, s15, s10, s1, s5]

Robert Pisarczyk and Jack Fitzsimons founded Oblivious in 2020, and both hold Oxford doctorates. RTE reports backgrounds across computer engineering, data science, machine learning and cryptography, and the company's About page places the founders' expertise in machine learning, security and privacy technologies.

Fitzsimons carries the externally visible research record. Silicon Republic quoted him saying Oblivious had worked with the UN Committee of Experts for two years as part of its Privacy-Preserving Technologies Task Team, Harvard's OpenDP published a use-case interview with him, and he co-authored the differential privacy auditing paper the 2026 Privacy Enhancing Technologies Symposium accepted.

The reviewed record documents no prior exit and names no executive beyond the two founders. RTE put the company at 13 staff when it announced the 2023 seed round, and no cited source updates that figure. [s2, s8, s11, s12, s13]

The homepage text carries ISO 27001, SOC 2 Type 1 and SOC 2 Type 2 labels. Nothing in the reviewed record backs them. A trust-surface probe found no DNS record for trust.oblivious.com or security.oblivious.com and a 404 at the /trust, /security and /compliance paths, so the labels stand as claims the vendor makes about itself.

The engineering evidence is public and checkable. The differential privacy auditing paper reports 13 privacy violations across 12 open-source libraries and its artifact carries the symposium's Available, Functional and Reproduced badges, and the company published the auditing tool as free software.

Legal identity is clear. The terms and conditions name Oblivious Software Limited at a Dublin address as the operator of the website and the services. An OpenCorporates registry record for the Ireland jurisdiction lists the same entity as company number 669274, a private company limited by shares with status Normal, at a Malahide, Co. Dublin address carrying the same postcode as the one in the terms, and gives 2025-09-30 as the date its last annual return was made up to. [s1, s19, s13, s14, s16, s7, s20]

Add analyzed competitors to compare them side by side with Oblivious.

Oblivious aims at organisations that hold regulated data and cannot use it. The homepage names financial institutions detokenising card data, banks aligning with PCI DSS 4.0, DORA and NIS2, engineering teams migrating high-risk workloads, and cross-border teams working under access restrictions.

The independent record describes clients by sector, apart from the United Nations, which it mentions by name. Tech.eu reported that Oblivious counts telecom companies, the United Nations, official statistics offices and AI startups among its clients, and Silicon Republic reported its work with national statistics offices through the UN PET Lab.

The stakeholders the pages address are data and privacy owners more than security teams. The pitch is written around unblocking analytics, reducing audit scope and satisfying GDPR, HIPAA and CCPA, and the AGENT console serves administrators, data owners and data scientists. No cited source names the budget holder.

The confidential computing capability is specific and documented. OBLV Deploy runs applications inside hardware secure enclaves with no code changes, a key management system releases decryption keys only after attestation confirms the running software and configuration, and a proxy validates that connections reach only attested environments.

The differential privacy capability is equally concrete. AGENT enforces differential privacy on query results, tracks a privacy budget per user and per team against every sensitive source, and reaches data scientists through Jupyter notebooks and SQL clients rather than a separate tool.

Where Oblivious goes beyond product claims is correctness. The company and its academic co-authors built a grey-box auditing method, found 13 privacy violations across 12 open-source differential privacy libraries including SmartNoise SDK, Opacus and Diffprivlib, and the 2026 Privacy Enhancing Technologies Symposium accepted the paper with its artifact marked Available, Functional and Reproduced. The company's blog puts the count at 11 libraries rather than the paper's 12, and both report the same 13 violations.

The motion is community-led. Oblivious runs the annual Eyes-Off Data Summit, operates the Antigranular competition platform, publishes a weekly privacy newsletter, and co-launched the PET Arena capture-the-flag competition with TikTok in December 2025.

Institutional partnerships do the credentialing that named customers usually do. Silicon Republic reported that Oblivious joined the UN PET Lab programme alongside OpenMined, and Harvard's OpenDP published a joint account of a UN trade-data pilot and of work with an unnamed major US telecom company.

What the funnel lacks is a named private-sector customer. Every buying signal in the reviewed record is a category, a partnership or an event, and across the reviewed commercial pages the visible conversion paths are a demo request and a shared mailbox.

No paid price appears in the reviewed record. The terms and conditions describe Free, Professional and Enterprise variants and say the Professional and Enterprise ones may need separate arrangements, and the homepage and product pages carry a demo request rather than a rate card.

What the pages do document is the enterprise packaging. AGENT builds on Antigranular by adding self-hosting, configurable libraries and authorisation flows, connectors to enterprise data sources, and privacy-budget administration per user and team.

For a buyer, that means procurement starts with a conversation. Nothing in the reviewed record lets a team size the cost of a pilot before contacting the vendor.

The customer runs the software. OBLV Deploy installs into the customer's Kubernetes clusters and keeps standard Kubernetes tooling intact, and AGENT self-hosts on the customer's own infrastructure with the libraries, privacy budgets and authorisation flows the customer configures.

The operational surface is real work. The documentation covers enclave lifecycle, manifests, outbound allowlists, certificate management including ACME issuance, key generation against an external key management system, telemetry through OpenTelemetry, Prometheus and Grafana Loki, and enclave pod autoscaling.

Cloud coverage is narrow today. The product page lists AWS availability with Azure in preview and more providers coming soon, so a customer standardised on another cloud is waiting.

The compliance posture is asserted rather than published. The homepage text carries ISO 27001, SOC 2 Type 1 and SOC 2 Type 2 labels, and a trust-surface probe found no DNS record for trust.oblivious.com or security.oblivious.com and a 404 at the /trust, /security and /compliance paths, so a buyer starts from a request.

The engineering evidence runs the other way. The auditing paper is public. The 2026 Privacy Enhancing Technologies Symposium accepted it and marked its artifact Available, Functional and Reproduced, and the company released dp-recorder as a free Python package rather than keeping the method to itself.

Legal identity is unambiguous. The terms and conditions name Oblivious Software Limited at a Dublin address as the operator of the website and services, which matches the About page and the funding coverage. The record is not only the vendor's own. OpenCorporates lists Oblivious Software Limited in the Ireland jurisdiction as company number 669274, type LTD - Private Company Limited by Shares, status Normal, and records its last annual return as made up to 2025-09-30.

Oblivious depends on other people's infrastructure by design. The enclave isolation comes from AWS Nitro Enclaves and equivalent cloud hardware, the orchestration is Kubernetes, and key release runs through the customer's own key management system.

The differential privacy stack is borrowed too, and the company says so. It told Harvard's OpenDP that the OpenDP library and SmartNoise had been the backbone of its disclosure controls, and its community platform integrates OpenDP, SmartNoise and DiffPrivLib.

The ecosystem it does own is social. The Eyes-Off Data Summit, the Antigranular platform, the UN PET Lab work and the PET Arena competition with TikTok and an ETH Zurich co-organiser put Oblivious at the centre of a research community, which is a distribution channel for a category that still has to explain itself.

Robert Pisarczyk and Jack Fitzsimons founded Oblivious in 2020 and both hold Oxford doctorates, with backgrounds RTE describes as computer engineering, data science, machine learning and cryptography. Pisarczyk is the chief executive.

Fitzsimons carries the public research record. Silicon Republic quoted him saying Oblivious had worked with the UN Committee of Experts for two years as part of its Privacy-Preserving Technologies Task Team, Harvard's OpenDP published an interview with him, and he co-authored the paper the 2026 Privacy Enhancing Technologies Symposium accepted.

The team is small and the record is old. RTE put the company at 13 staff when it announced the April 2023 seed round and reported a plan to add seven more, and no cited source updates that count.