The Difference Between Knowing and Being Allowed

Capitolis Raises $220M for eSecLending Acquisition

Capitolis is raising money from the same kind of institutions its technology is supposed to make more efficient.

That makes its new $220M financing more interesting than the headline number. Citi led a $120M Series E at a $1.9B valuation. Bank of America, Nomura, and Tradeweb joined as new strategic investors, while Barclays, BNP Paribas, J.P. Morgan, State Street, and UBS returned. The remaining financing is roughly $100M of debt.

The mix is deliberate because Capitolis has already picked the next piece of infrastructure it wants to own. The company agreed to acquire eSecLending for $200M in cash, subject to closing conditions and regulatory approval.

eSecLending brings 26 years of securities-lending experience and relationships with pension funds, insurers, asset managers, banks, and prime brokers. Capitolis already helps institutions address funding, capital, liquidity, and balance-sheet pressure through its Capital Marketplace and Portfolio Optimization businesses.

Put those networks together and the strategy gets larger than another product module. Banks need inventory and financing. Asset owners have securities to lend. Both sides need reliable data, counterparties, controls, and execution. Capitolis is trying to own more of that handoff.

CEO and founder Gil Mandelzis built Traiana before leading EBS BrokerTec, so this is familiar territory: regulated markets rarely become infrastructure because somebody ships a clever interface. They become infrastructure when enough institutions trust the same system to carry meaningful work.

That is why the cap table matters. The banks are investors, potential users, distribution partners, and demanding judges of whether Capitolis can make financial resources move more efficiently. Tradeweb adds another market network and a direct thesis around electronifying securities lending.

CTO Murugan Manickam arrived in September after 17 years at Bank of America to lead global engineering and Capitolis' data and AI strategy. The company is pairing a broader institutional network with leadership that has lived inside the systems it now has to connect.

The acquisition has not closed, and integration will decide whether this becomes a stronger platform or an expensive collection of adjacent capabilities. Still, the financing shows what Capitolis believes it needs next: a live network of asset owners that can meet its bank network on the other side of the trade.

Deeper Analysis 👇 https://devcuration.com/articles/capitolis-raises-220m-for-esec-lending-acquisition

Zeroset Raises $5.2M for Enterprise Agent Memory

Every enterprise has a second operating system. It lives in exceptions, handoffs, approvals, stale tickets, and the quiet reason somebody broke the process last Tuesday.

The software records what happened. The people remember why. AI agents usually arrive somewhere between those two realities and discover that “complete the workflow” is doing Olympic-level work as a sentence.

Zeroset has raised a $5.2M pre-seed round to build for that gap. Gradient and 2048 Ventures co-led the financing, with Leblon Capital participating. The San Francisco company is led by co-founders Akshat Kannan, CEO, and William Zhang.

Their first product, Nebula, is designed to capture the state and traces created as people and agents work across tickets, messages, approvals, handoffs, tool calls, and decision records. Instead of reconstructing the company from documents every time it wakes up, an agent can query what changed, what remains open, and which prior decisions govern the next move.

That distinction matters because enterprise work is rarely one clean task. A product request can move from Slack to a ticket, inherit a verbal approval, hit a policy exception, change owners, and arrive at release with half its logic living in somebody's memory. A model can be brilliant at reasoning and still be wrong about the state it is reasoning from.

Zeroset is borrowing a useful idea from robotics: capable policies need a representation of the environment in which they act. The company's version is an enterprise world model, built from workflow traces and designed to remain portable across frontier and open models. The operating knowledge belongs to the enterprise, not whichever model provider happens to be fashionable this quarter.

Nebula is still in closed research preview. Gradient and 2048 Ventures report that it beat comparison systems on LongMemEval retrieval accuracy by 20%, with sub-50ms median retrieval and lower token use. Those are company and investor claims, not independent proof that the system can already carry a complex production workflow without supervision.

The bill for proving it will not be theoretical. Business Insider reported that one peak training run cost roughly $100K over two weeks. Zeroset plans to use the round to hire researchers and systems engineers, support early enterprise deployments, and pay for more model training.

The enterprise-agent market has spent plenty of money making models more capable. Zeroset is betting the next layer of value comes from making the company legible to those models, one changing decision at a time.

Guardrail Technologies Closes Nearly $3M in Seed Funding

The AI agent with access to payroll, source code, customer files, and cloud infrastructure does not need a badge. It needs permissions. That small change turns an exciting automation project into a very familiar security problem moving at unfamiliar speed.

Guardrail Technologies, Inc. has now closed two Seed rounds totaling nearly $3M, led by Filter Fund Venture Capital with participation from unnamed strategic angels. This is cumulative funding across two rounds, not one fresh $3M check.

Founder and CEO T.J. Marlin spent more than 20 years in forensic technology, including leading EY's global forensic technology work. Investigators are usually called after the incident, when the company needs to know who acted, what changed, which controls failed, and whether the evidence can survive scrutiny. Guardrail is trying to build those questions into the operating layer before the post-mortem begins.

Its AI Traffic Light product evaluates AI-generated and human-written code and returns a Green, Amber, or Red signal. AI Command Center is designed to give enterprises a central view of agent behavior, policies, integrations, performance, and audit records. The company reports more than 1M AI security scans, three issued patents, and operations on two continents.

The outside pressure is getting less theoretical. NIST has warned that agentic AI needs stronger identity and authorization because agents can reach tools, data, and credentials. OWASP traces excessive-agency risk to too much functionality, permission, or autonomy. Enterprise AI is becoming a workforce with machine-speed access while many organizations still govern it like software that only answers questions.

That distinction is where enterprise buying decisions will form. A clever agent earns attention. An agent that can explain its authority and leave evidence earns access to real systems.

Filter Fund founder Richard K. Sussman also serves as Guardrail's executive chairman. The financing will support expansion into Asia and Europe after Guardrail established Australian operations, plus hiring across engineering, AI/ML, infrastructure, customer success, marketing, and enterprise sales.

Nearly $3M is not a giant AI round. It is enough to test whether Guardrail can make control, evidence, and accountability travel with the agent instead of arriving after the damage. As more software gets the authority to act, the companies selling that authority will share the table with the ones responsible for containing it.

Deeper Analysis 👇 https://devcuration.com/articles/guardrail-technologies-closes-nearly-3m-in-seed-funding

SAP to Acquire TechWolf for AI Workforce Intelligence

An enterprise can automate a workflow long before it can explain who actually does the work, which skills keep it moving, and where the process changes when the org chart stops paying attention.

That gap is what SAP is buying with TechWolf.

SAP has signed an agreement to acquire the Ghent-based work-intelligence company and make its “context graph for work” an intelligent core of SAP SuccessFactors. The price and transaction terms were not disclosed. The deal is expected to close in Q4 2026, subject to customary conditions including regulatory approval.

TechWolf connects to HR and operating systems, then models work at 3 levels: tasks inside a job, skills people apply, and the external labor market. That evidence supports planning without pretending a job title tells the whole truth.

For SAP, the acquisition reaches beyond HR software. AI agents need grounding. A system can know the approved process and still misunderstand who handles the exception, which capability is scarce, or how a role changed after the documentation froze. TechWolf gives SAP a stronger map of that moving organization.

Andreas De Neve, CEO and co-founder, and Jeroen Van Hautte, CTO and co-founder, started TechWolf in 2018 with Mikaël Wornoo after a university project in Ghent. The company later raised a €10M Series A and a $42.75M Series B with SAP among the backers.

TechWolf says its U.S. ARR grew from $1M to $15M over 18 months. That is company-reported, but the choice is revealing. Demand was outrunning the company's ability to serve it alone, while SAP already had the distribution and operational systems where evidence about work lives.

TechWolf is expected to remain independent under Andreas De Neve in Ghent, keep offices in London and New York, add San Francisco, and serve SAP and non-SAP customers. That openness is part of the asset and one of the integration tests.

SAP has spent years assembling context around enterprise change. WalkMe added digital-adoption signals. SmartRecruiters expanded talent acquisition. TechWolf adds a living model of tasks and skills that can make Joule and workforce agents harder to fool with stale assumptions.

The operating question now moves inside SAP: can it preserve TechWolf's independent evidence layer while embedding it deeply enough to change hiring, reskilling, role design, and the agents asked to act on all 3? The org chart will not be enough to settle that argument.

Deeper Analysis 👇 https://devcuration.com/articles/sap-to-acquire-techwolf-ai-workforce-intelligence

Hilt Raises $4.2M Seed for Data Movement Governance

Before a security team cuts off a workstation, somebody has to explain what its owner was doing. Hilt is building that explanation around how data moves, including the moves an employee, vendor or AI agent was allowed to make.

The company announced a $4.2M Seed round led by Array Ventures on October 7. Verdict Capital, Base10 Partners, Brickyard, Liquid 2 Ventures, Sequel, Sarah Smith Fund and Alumni Ventures participated. Reported total funding is $4.7M, including an earlier $500K pre-seed led by Pear VC.

William Cielen, Hilt's cofounder and CEO, found the initial pull in high-frequency trading. A customer needed security that could observe sensitive systems without becoming another delay inside them. Cielen's own timeline places the pivot in January 2026, after starting Hilt in San Francisco the previous September.

CTO and cofounder Alexandre Genest and founding engineer Zin Bitar are working on a product that observes activity at the kernel. Hilt describes collectors that gather metadata, attach identity and workload context, identify unusual movement and give the control plane a way to isolate a host at the network during an investigation.

That last step changes the conversation. Flagging a transfer asks an analyst to investigate. Isolating a host can interrupt the person doing the job. A buyer needs to understand the evidence behind the intervention, the systems it covers and the work it might stop.

Hilt's deployment model puts raw telemetry inside the customer's environment. Its FAQ also discloses an anonymized aggregate signal sent back to improve detection. That detail belongs in the buying conversation, alongside the distinction between metadata-only collection by default and optional content inspection.

The team has moved to Chattanooga, Tennessee, with investor Brickyard. Cielen says the financing will primarily support hiring and building a market presence. For a young security company, those hires carry the work of making a technically demanding deployment understandable to the next customer.

The commercial opportunity sits in that handoff: from a pattern in the telemetry to a decision a security team can defend. Hilt has described how it wants the system to work; public independent detection benchmarks and named customer outcomes remain scarce. The people evaluating it will still need to put their own normal working day in front of the collector and see which actions deserve an interruption.

Deeper Analysis 👇 https://devcuration.com/articles/hilt-raises-4-2m-seed-data-movement-governance

Medical Microinstruments Raises $125M for Symani Expansion

Massimiliano Simi started with the wrist. Miniaturizing an articulated surgical instrument gave Medical Microinstruments a way to carry a surgeon's movements into anatomy measured in fractions of a millimeter. A company grew around that instrument.

Simi, now CTO, founded MMI in 2015 with Giuseppe Prisco and Hannah Teichmann near Pisa, Italy. Jacksonville-based MMI announced a $125M Series D on October 8, led by existing investor BioStar Capital, with S3 Ventures, Angelini Ventures, an undisclosed strategic corporate investor and existing backers participating.

CEO Mark Toland is taking Symani into its next commercial stretch. MMI reports approximately 50 installed systems and 3,500 procedures worldwide. Those counts mean persuading a hospital to build a microsurgery program around a new way of working.

Symani's NanoWrist instruments combine articulated movement, tremor reduction and motion scaling. The surgeon stays in control. The technology supports delicate work on small vessels and lymphatic ducts, with authorized applications that vary by region. MMI is financing U.S. and global expansion alongside clinical trials and outcomes data.

A hospital buys the support surrounding that precision, too. MMI offers simulation, hands-on training, onsite clinical assistance and a service program for maintenance and repairs. An operating room needs people who can use the equipment and somebody available when the equipment needs attention. That brings the instrument into everyday care.

BioStar has backed MMI before. Its return as lead investor puts fresh capital behind that adoption effort, while the company develops energy instruments and digital surgery capabilities. MMI's current investor page puts total investment to date at approximately $345M, following a $110M Series C led by Fidelity Management & Research Company in 2024.

The research reaches further. REMIND is studying robotic microsurgery in Alzheimer's patients with lymphatic abnormalities. The 15-person early feasibility study examines safety and feasibility; this brain-disease application remains investigational, with safety and effectiveness unestablished. Hospitals evaluating Symani's current surgical uses and researchers investigating a new application are working on different calendars.

Simi's miniature wrist now comes with a growing clinical organization around it. As Toland expands the footprint, each hospital joining that network brings another team to train, another program to support and more cases to inform where microsurgery goes.

Deeper Analysis 👇 https://devcuration.com/articles/medical-microinstruments-125m-series-d-symani

Voltus Raises $225M Series D to Expand Grid Flexibility

Google is financing a power plant assembled from other people's batteries and thermostats. Voltus has to make those individual customers willing participants in a capacity plan large enough for a hyperscaler.

That relationship sits inside Voltus's $225M Series D, announced October 8. Generation Investment Management led the oversubscribed round, with Activate Capital and Vitol as co-leads. Broadscale Group, Climate Investment, NGP, Ajax Strategies and All Aboard Fund also participated.

The San Francisco company connects flexible electricity use and distributed energy resources to power markets. A business can reduce consumption, shift it, or supply power from an eligible asset, then earn money for helping the grid. Coordinating those decisions gives electricity buyers another source of capacity while participating customers keep running their own lives and businesses.

Google's 3-year agreement with Voltus, announced in June, aims to unlock up to 100 MW in PJM. Google funds the virtual power plant; Voltus recruits and orchestrates resources at local homes and businesses and pays participants. The data center's appetite for electricity becomes a reason to invest in the community around it.

CEO and co-founder Dana Guernsey brings experience from EnerNOC, Ambri and FirstFuel. President Matthew Plante and Chief Architect Neil Lakin are fellow co-founders, alongside original co-founder Gregg Dixon. This is a company founded in 2016, with years of customer and electricity-market work behind the new financing.

The team has also moved closer to the equipment. Voltus acquired Brightfield AI in June, bringing in battery-development capabilities and CTO Larsh Johnson, previously CTO at Stem. Helping a customer assess a battery project extends the relationship beyond asking what an existing site can already contribute.

Voltus reports 8.5 GW of distributed resources and more than 22,000 participating sites. Its 20 GW ambition for 2030 is a target, with this capital supporting Bring Your Own Capacity programs in PJM and MISO, more connected batteries, and expansion across the U.S. and Canada.

The commercial work lives in the individual agreement: what a site can offer, when it can respond, how that response fits the market, and what the participant earns. A cooling system belongs to somebody who still needs the building comfortable. That person's permission travels all the way into the capacity a large electricity buyer is paying Voltus to organize.

Deeper Analysis 👇 https://devcuration.com/articles/voltus-raises-225m-series-d-grid-flexibility

Alight Acquires Abett for Benefits Data and Engagement

For Mike Hanlon, Alight brings a distribution opportunity measured in decisions. The Abett founder points to the millions of people an administrator can reach while they are actively choosing benefits.

That is valuable territory for a company connecting the information behind those choices. An employer can buy medical coverage, pharmacy services and specialist programs from different vendors, then have to work out how the pieces serve the same person.

Alight Solutions announced its acquisition of Abett on October 1, 2026. Terms were undisclosed. Rohit Verma, Alight's CEO, is bringing benefits data infrastructure into a business that already handles enrollment, customer care, navigation and ongoing service events.

Abett's Data Engine integrates information from across the benefits ecosystem, supports member-level engagement and measures program outcomes. Inputs include claims, clinical records, HR data and specialist services. Those records help a benefits team investigate how its vendors work together.

Alight Worklife supplies another part of the relationship: the interface through which people encounter benefits and receive guidance. Pairing that access with Abett's infrastructure creates more opportunities to connect an identified need with a service while the employee is already paying attention.

Hanlon describes the combination as a way to understand need, take action and measure change. Verma wants to extend those capabilities to existing clients and beyond. The ambition follows the employee through a decision, rather than ending when an employer receives another report.

Naveen Baweja, Alight's CTO, leads the buyer's technology organization. Combining data and participant services puts practical work around that organization: maintaining vendor connections, carrying the right information into the experience and making the resulting guidance useful to the person receiving it.

Benefits leaders will also have a sharper purchasing conversation. When an administrator can help direct people toward services and measure their use, employers need to understand how that measurement works across the vendors they have chosen. Compatibility and usable evidence belong in the same discussion as the employee experience.

The announcement leaves pricing, integration timing and measured improvements open. Hanlon has put Abett closer to the places where benefits decisions happen; the employer still has to connect that opportunity to the programs its people will actually use.

Deeper Analysis 👇 https://devcuration.com/articles/alight-acquires-abett-for-benefits-data-and-engagement

micro1 Commits $1B to Enterprise Data, Backed by Financing

An SOP written to spare the next hire a messy handoff has acquired a potential buyer. micro1 is putting a procurement budget behind the knowledge companies usually keep inside the business.

On October 9, the data lab committed to spending $1B over the next 12 months acquiring and licensing enterprise operational data. micro1 says Citi and Hercules Capital are providing capital to finance the initiative. The financing amount and terms remain undisclosed; $1B describes planned spending.

Founder and CEO Ali Ansari is extending a business built around human expertise into the records organizations accumulate while doing their jobs. A process document can explain the steps. Project history can preserve decisions and exceptions.

That context has a customer. micro1 uses de-identified business data to construct reinforcement-learning environments where models and agents practice operating through real workflows. The company plans to expand the industries and kinds of work represented in those environments.

Its production stack already connects several jobs: Zara recruits and vets domain experts, Flow supports dataset creation and review, and Merit tracks pipeline performance. Cortex brings expert judgment into evaluation and improvement of agents. Buying enterprise records gives that machinery another input to work with.

For a company considering participation, the conversation quickly becomes commercial and operational. What can be licensed? Who approves it? Which details carry useful context, and which details must stay private? A valuable dataset still needs someone authorized to explain the boundaries around it.

micro1 says participating companies keep ownership of the underlying data. Its program describes agreed scope, redaction, limited access, sample review and retention controls. Those are company-stated practices, with the actual partnership negotiated individually. A neat privacy paragraph cannot sign the agreement for either side.

The interesting supplier here may be a business that never thought of itself as an AI company. Its working knowledge was produced to serve customers, reconcile accounts, handle projects or keep operations moving. Licensing gives that same work another commercial audience.

Citi and Hercules Capital enter the story at the acquisition budget. The organizations supplying the data enter it at the permission boundary. Somewhere between those conversations, micro1 has to turn an account of how work happened into an environment where an agent can learn to do it.

Deeper Analysis 👇 https://devcuration.com/articles/micro1-commits-1b-enterprise-data-financing

PsiQuantum Gets Conditional $150M Quantum Loan Commitment

Victor Peng has spent a career getting complicated chips into the world. At PsiQuantum, the equipment around those chips is becoming a federal financing priority.

The former AMD president and Xilinx CEO now leads a company with a conditional loan commitment of up to $150M from the U.S. Department of War's Office of Strategic Capital. Announced October 8, the proposed financing would expand manufacturing and prototyping at its 140,000-square-foot Milpitas hub, alongside company co-investment.

The purchases are wonderfully unglamorous: equipment to test components and assemblies, infrastructure to put systems together, and advanced cryogenics. Anyone hoping quantum computing would skip the machinery has been handed a fairly substantial shopping list.

PsiQuantum's 4 co-founders remain central to that work. Jeremy O'Brien is executive chairman, Mark Thompson is CTO, Terry Rudolph is chief architect, and Pete Shadbolt is chief scientific officer. Their photonic approach puts particles of light to work as qubits and builds around semiconductor manufacturing, telecom fiber and modular cooling.

That choice gives the business a relationship with an existing industrial base. GlobalFoundries manufactures its quantum photonic chips in New York. PsiQuantum says it spent approximately $200M with hundreds of American suppliers across 38 states in 2025. A quantum project reaches quite a few loading docks before it reaches an application.

Private investors have financed the ambition, too. The company's $1B Series E in September 2025 was led by BlackRock-affiliated funds and accounts, Temasek and Baillie Gifford. This latest commitment addresses equipment capacity with debt, subject to financial, legal, technical and regulatory conditions. The loan is not guaranteed or evidence that cash has arrived.

Federal support has distinct jobs. A $125M DARPA agreement announced in July funds testing and evaluation. A $100M Commerce award finalized in September supports work on switches, detectors and packaging. Adding those figures into a new capital-raised headline would lose the engineering purposes and the obligations attached to each.

DARPA's current Stage C includes PsiQuantum and 5 other companies pursuing different approaches. The evaluation follows the hardware as it develops, while PsiQuantum prepares sites in Chicago and Queensland for its planned utility-scale machines. Milpitas is where more of the parts, cabinets and measurements needed for those sites would come together.

Deeper Analysis 👇 https://devcuration.com/articles/psiquantum-conditional-150m-loan-quantum-manufacturing

Catalyst Raises $30M Seed for Retail AI Trading Agents

Dylan Iskandar and Justin Zheng are asking retail investors to describe a trade before choosing its machinery. At Catalyst, the idea goes in as plain English, then becomes a strategy the customer can inspect, authorize and leave running inside chosen limits.

Sequoia Capital led Catalyst's $30M Seed round, announced October 8, with Jump Trading, PeakXV, Lux Capital, AntiFund, Coinbase and Premji Invest participating. That is a sizeable early commitment to a product trying to make financial execution understandable to a wider retail audience.

The co-founders are working on a gap a faster order button leaves open. Someone can have a view about a market and still struggle to choose the instrument, define the trigger or understand how several positions interact. Getting access to an asset leaves plenty of decisions between the opinion and the order.

Catalyst builds a visual workflow from that intent. Its official materials describe agents monitoring market data, news, social activity and sentiment, then executing actions under the user's parameters. The supported markets include spot tokens, perpetuals, tokenized stocks and prediction markets. Sequoia also describes backtesting as part of the path to deployment.

That puts an unusually personal job inside the interface: helping someone recognize their own instruction after software has translated it into financial exposure. A sentence can sound perfectly reasonable while leaving its budget, timing or exit conditions unresolved. Those details belong where the customer is deciding what to allow.

Catalyst says users set budgets and execution conditions, can pause or modify strategies, and retain custody of their funds. Those are company-described controls. The service can act automatically within its boundaries, so an approved strategy may keep making decisions after the conversation ends.

The research preview is free, and the website still invites people onto a waitlist. No commercial rollout date is specified. Broader retail access is the stated direction. That gives Justin Zheng and Dylan Iskandar an adoption problem with a very concrete shape: explain the strategy well enough that somebody can decide how much responsibility to hand it.

As Catalyst reaches more ordinary investors, the pause and modification controls will sit beside the promise of automated execution. A customer whose view has changed needs that change to reach the workflow before the next market signal becomes another trade.

Deeper Analysis 👇 https://devcuration.com/articles/catalyst-30m-seed-ai-trading-agents

TypeSafe AI Raises $870M Series A for Machine-Native AI

A confidence score gives a software team another decision to own. TypeSafe AI is building around the point where a model's judgment enters somebody else's application and a developer decides what happens with it.

CEO Diogo Almeida, COO Sasha Sheng and CTO Erik Gafni announced an $870M Series A at a $7.5B valuation on October 9, 2026. Andreessen Horowitz led the round, with Sequoia Capital, returning investor DCVC and unnamed angels participating. Martin Casado is joining the board.

That follows the $40M Seed led by DCVC in September. The San Francisco company, founded in 2024, now has approximately $910M in disclosed funding across those 2 rounds. The new announcement leaves the valuation basis and any secondary component unspecified.

Jev gives software typed answers it can use directly. A developer supplies the state and a question; the model returns a choice, score or probability. Choice and Score include confidence, giving the application a way to decide whether to act, ask another question or hand the judgment back to a person.

The interesting design choice is in TypeSafe's documentation. Break a complicated judgment into smaller questions, evaluate the factors separately, then combine them in code. When priorities change, a developer can change a weight in that code instead of trying to persuade a prompt to behave differently.

That keeps an important conversation inside the team building the product. How much should urgency count? Where should uncertainty stop an action? The software owner can inspect those choices, revise them and decide which decisions belong in the background.

Almeida brings AI research experience; Sheng comes from Meta/FAIR; Gafni has built production AI systems. Their work now sits between a model lab's ambition and a customer's need to keep an application intelligible to the people operating it.

TypeSafe plans more machine-native models and enterprise features. The investor group is financing a business where the buyer may care deeply about intelligence while spending very little time looking at its output. A returned value has to justify its place in a workflow through the decisions it improves.

For the team integrating Jev, the next design conversation starts with the confidence threshold. A product manager's judgment about acceptable uncertainty can become an executable rule, and another team will eventually have to understand why that rule was chosen.

Deeper Analysis 👇 https://devcuration.com/articles/typesafe-ai-raises-870m-series-a-machine-native-ai