Kyndryl’s Ismail Amla discusses the company’s new policy as code process, and how it can help address AI issues such as agentic drift.

When it comes to AI adoption in enterprise, compliance concerns are becoming ever more important.

According to Kyndryl’s most recent Readiness Report, 31pc of enterprise customers cite regulatory or compliance concerns as a primary barrier limiting their organisation’s ability to scale recent technology investments.

2026 marks an important point on the AI compliance timeline in particular, with the EU’s AI Act transparency rules coming into effect in August.

Last month, Kyndryl announced its new ‘policy as code capability’ – a new process designed for creating policy-governed agentic AI workflows for enterprises.

“Policy as code is the process of translating an organisation’s rules, policies and compliance requirements into machine-readable code, so AI systems are restricted to only operating within pre-defined guardrails,” explains Ismail Amla, senior vice-president at Kyndryl Consult. “Human experts continue to oversee all activities related to these processes.”

Compliant design

“Many organisations, especially those in complex, highly regulated environments, want to scale agentic AI, but are held back by concerns around security, compliance and control”, says Amla.

Speaking to SiliconRepublic.com, he says policy as code can help organisations support “consistent policy interpretations” and define clear operational boundaries, subsequently ensuring agent actions are explainable, reviewable and “aligned with organisational standards”.

Amla also says the framework can help reduce costs, accelerate decision-making, eliminate errors and “power AI-native workflows within defined policy guardrails”.

“By embedding policy and regulatory requirements directly into AI agent operations, policy as code can help organisations execute AI workflows that are governed, transparent, explainable and aligned to business requirements.”

But what about the long-term applications of policy as code?

Amla says the main benefit of the process is “trust through stronger governance, better transparency, lower operational risk and more reliable AI at scale”.

“Managing agentic workflow execution in this way supports controlled and responsible deployment of policy-constrained AI agents in sectors such as financial operations, public services, supply chains and other mission-critical domains, where reliability and predictability are essential,” he explains.

Catch the drift

Over the past year, according to Amla, the biggest change he’s noticed in AI adoption is that organisations are moving beyond proofs of concept and “focusing more seriously on what it takes to make AI work in production and at scale”.

“That means more attention on infrastructure, governance, data quality and organisational readiness,” he says. “Organisations are moving from experimentation to making more strategic decisions with the experience they have gained to drive higher value outcomes and performance for their organisation, and receive a return on their investment.”

But with increased focus on serious AI integrations comes risk, particularly if an organisation is not fully prepared.

Amla warns of something called ‘agentic drift’, which refers to when an AI agent can appear reliable while working toward unwanted outcomes due to a gradual separation from the agent operator’s original intention or goal.

“Agentic drift creates pressing challenges for all organisations, but it is especially acute in the public sector and highly regulated sectors, such as banking and healthcare,” says Amla.

“In these industries, organisations cannot move from pilots to production if issues around control, trust and compliance remain unresolved. It’s clear enterprises urgently need a way to constrain what agents can do at runtime and close governance gaps long before drift leads to financial or compliance failures.”

Amla believes that policy as code can help address this issue, due to its ability to allow businesses to translate their rules and policy into machine-readable instructions that “govern how AI agents reason, adapt and act”.

“This greatly reduces the risk of agentic drift,” he says. “It also alleviates the trust and compliance concerns standing between large enterprises and a return on their AI investments.”

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

Sony is shutting down Dark Outlaw Games, a first-party game studio led by former Call of Duty producer Jason Blundell, Bloomberg‘s Jason Schreier reports. Before leading Dark Outlaw Games, Blundell was the head of Deviation Games, which was an independent studio, but also happened to be developing a PlayStation game before it shut down, Schreier says.

Dark Outlaw Games had yet to announce what it was working on, but considering Blundell’s experience with the Call of Duty franchise, it seems likely the studio was developing a multiplayer project for PlayStation. Blundell was a programmer and producer at Activision before making the jump to Treyarch to work on Call of Duty 3, and he contributed to multiple Call of Duty: Black Ops games after that, including serving as the director for the campaign and Zombies mode of Call of Duty: Black Ops III and the career and Zombies modes of Call of Duty: Black Ops 4.

Engadget has contacted Sony for more information about the fate of Dark Outlaw Games. We’ll update this article if we hear back.

The studio’s shutdown is being paired with cuts to staff at PlayStation focused on mobile development, according to Schreier. Sony has made a habit of laying off staff and shutting down studios in the last year, seemingly as a way to retreat from an earlier investment in online, live-service multiplayer games. The company shut down Bluepoint Games in February following attempts to get a live-service God of War game off the ground. Sony also closed Firewalk Studios after the spectacular failure of multiplayer shooter Concord in October 2024. And a year before that, Naughty Dog officially abandoned work on a standalone multiplayer version of The Last of Us in December 2023.

That leaves Sony with at least two Horizon Zero Dawn spin-offs, a co-op game from original developer Guerilla Games and a MMO from developer NCSoft; Fairgame$, which is still in active development despite the departure of Haven Studios head Jade Raymond; Arrowhead Game Studios’ Helldivers 2; Bungie’s Destiny 2 and Marathon; and if you really want to stretch, Gran Turismo 7. Sony clearly hasn’t given up on producing online multiplayer games, but it’s not hard to characterize its attempt to expand into the space as a disaster.

For developers using AI, “vibe coding” right now comes down to babysitting every action or risking letting the model run unchecked. Anthropic says its latest update to Claude aims to eliminate that choice by letting the AI decide which actions are safe to take on its own — with some limits.  

The move reflects a broader shift across the industry, as AI tools are increasingly designed to act without waiting for human approval. The challenge is balancing speed with control: too many guardrails slows things down, while too few can make systems risky and unpredictable. Anthropic’s new “auto mode,” now in research preview — meaning it’s available for testing but not yet a finished product — is its latest attempt to thread that needle. 

Auto mode uses AI safeguards to review each action before it runs, checking for risky behavior the user didn’t request and for signs of prompt injection — a type of attack where malicious instructions are hidden in content that the AI is processing, causing it to take unintended actions. Any safe actions will proceed automatically, while the risky ones get blocked.

It’s essentially an extension of Claude Code’s existing “dangerously-skip-permissions” command, which hands all decision-making to the AI, but with a safety layer added on top.

The feature builds on a wave of autonomous coding tools from companies like GitHub and OpenAI, which can execute tasks on a developer’s behalf. But it takes it a step further by shifting the decision of when to ask for permission from the user to the AI itself. 

Anthropic hasn’t detailed the specific criteria its safety layer uses to distinguish safe actions from risky ones — something developers will likely want to understand better before adopting the feature widely. (TechCrunch has reached out to the company for more information on this front.)

Auto mode comes off the back of Anthropic’s launch of Claude Code Review, its automatic code reviewer designed to catch bugs before they hit the codebase, and Dispatch for Cowork, which allows users to send tasks to AI agents to handle work on their behalf.  

Auto mode will roll out to Enterprise and API users in the coming days. The company says it currently only works with Claude Sonnet 4.6 and Opus 4.6, and recommends using the new feature in “isolated environments” — sandboxed setups that are kept separate from production systems, limiting the potential damage if something goes wrong.

A few months after its initial launch, Amazon has recently unveiled the Kindle Scribe Colorsoft in a brand new fetching Fig shade that’s proved especially popular.

In fact, the Fig-colour Kindle Scribe Colorsoft is so popular that it’s becoming increasingly difficult to get our hands on the e-reader, with shipping delays stretching well beyond the typical delivery windows we’d expect from Amazon.

At the time of writing, new orders for the Fig iteration in the US are expected to arrive anywhere between mid-April to mid-May. However, you can get your hands on the standard Graphite finish which is currently still in stock within the US. This suggests that the issue really only affects the newer colour option, rather than the entire product line.

Such differences in availability often point to supply constraints or production adjustments, particularly when a new finish launches after the initial release and demand shifts toward the latest variant.

Advertisement

Advertisement

It’s worth noting that at the time of writing, neither the Fig nor Graphite Kindle Scribe Colorsoft has officially launched in the UK. In addition, neither iterations are even available to pre-order, as the product page just states the e-reader is “coming soon”. Instead, you can opt into receiving an email to get notified on when the product will be available to buy.

Delays highlight uneven availability

The Kindle Scribe Colorsoft was initially only available in a Graphite option until Amazon recently introduced the new Fig finish, which seemingly appears to have drawn a considerably higher demand than anticipated. Either that, or the Fig shade has encountered production challenges soon after release.

However, delays tied to a specific colour variant are not uncommon, as sometimes manufacturing complexity or material sourcing can affect certain finishes differently than standard models.

In addition, the extended wait times also suggest that supply has not yet caught up with demand, especially as colour e-paper devices remain a relatively new category with more limited production scale compared to traditional e-readers.

Advertisement

Essentially, customers are left choosing between faster delivery by opting for the Graphite version, or waiting considerably longer to nab the Fig iteration instead.

Advertisement

This situation leaves buyers choosing between faster delivery with the Graphite version or waiting longer to secure the Fig model.

Same hardware, different buying experience

Following on from the above, it’s worth noting that both versions of the Kindle Scribe Colorsoft share the same core hardware, including an 11-inch colour e-paper display based on Kaleido 3 technology, which combines standard black-and-white clarity with lower-resolution colour output.

The device also integrates a redesigned front-light system and a textured display surface that improves writing feel, placing it closer to digital notebooks than traditional e-readers focused only on reading.

Advertisement

Storage options and connectivity remain consistent across variants, with support for Wi-Fi, Bluetooth audio, and bundled stylus input, which reinforces that the delay relates to availability rather than product capability.

Amazon has not provided a detailed explanation for the extended shipping times on the Fig model, but current delivery estimates suggest that availability may stabilise later in the Spring.

Advertisement

If you are exploring other options, our Best Kindle 2026 roundup highlights the top-performing e-readers available today.

Advertisement

If you haven’t watched Project Hail Mary yet, you should. Try to watch it on the largest screen possible. It’s beautiful, heart-warming and fun for any audience. I’ve been obsessed with it since I listened to the Audiobook with Ray Porter, and the cinema version doesn’t disappoint.

Movies like this help inspire people to be scientists and explorers, and to look for the good in people. It shows that no matter who you are, you can change the world.

That’s my mini review, but not the real reason for this article. Project Hail Mary has done something that makes me, a 3D-printing maker, happy, happy, happy. If you visit the Project Hail Mary website and look in the bottom-right corner, you can download a 3D model of a stylized spaceman used in the movie.

I’ll try not to spoil anything, but the little spaceman is given to the main character, Ryland Grace, to help him visualize the ideas that his companion is trying to portray. It’s a beautiful little model, and not the first time a company has done something like this to promote a movie.

Many years ago, Paramount released a 3D model from Transformers: Rise of the Beasts. This year, Markiplier created a haunted 3D-printed popcorn bucket that you could actually take to the theater and get a free popcorn order. It was gross and cool at the same time, but unique enough that a lot of people enjoyed making it.

My hope is that more movie studios will realize how well these files are received by the maker community and keep giving us more. A lot of the models VFX designers create can be converted into 3D-printable models with ease, and in the case of Project Hail Mary, this file was almost certainly a 3D-printed prop anyway. They have the file, so why not share it with the world?

While we didn’t have any Xenonite around to 3D print with, we did have some lovely silver silk PLA to make this fancy little spaceman. Printing it on the fantastic Bambu Lab H2D was a breeze with some supports as needed. The pattern of the model makes it look so surreal and gives it an alien quality that really makes it stand out. Print time was around four hours using PLA.

My next project after this is to print the same model in Iron filament from Protpasta and let it rust to really make it feel otherworldly.

Project Hail Mary is something of a cultural phenomenon right now, and rightfully so. Adding the ability to 3D models directly from the studio has added a little more advertising from a group of people who are very likely to love a deep sci-fi movie and share what they’ve made with the world. Let’s hope more movie studios see how successful this is and jump on the idea, too.

from the making-being-awful-more-efficient dept

All the people who have always brushed off concerns about surveillance tech, please come get your kids. And then let someone else raise them.

Lots of people are fine with mass surveillance because they believe the horseshit spewed by the immediate beneficiaries of this tech: law enforcement agencies that claim every encroachment on your rights might (MIGHT!) lead to the arrest of a dangerous criminal.

Running neck and neck with government surveillance state enthusiasts are extremely wealthy Americans. When they’re not adding new levels of surveillance to the businesses they own, they’re scattering cameras all around their gated communities and giving cops unfettered access to any images these cameras record.

Here’s how it plays out at the ground level: parents can’t get their kids enrolled in the nearest school because of surveillance tech. In one recent case, license plate reader data was used to deny enrollment because the data collected claimed the parent didn’t actually reside in the school district.

Just over a year ago, Thalía Sánchez became the proud owner of a home in Alsip. She decided to leave the bustle of the city for a quiet neighborhood setting and the best possible education for her daughter.

However, to this day, despite providing all required paperwork including her driver’s license, utility bills, vehicle registration, and mortgage statement, the Alsip Hazelgreen Oak Lawn School District 126 has repeatedly denied her daughter’s enrollment.

Why would the district do this? Well, it’s apparently because it has decided to trust the determinations made by its surveillance tech partner, rather than documents actually seen in person by the people making these determinations.

According to the school district, her daughter’s new student enrollment form was denied due to “license plate recognition software showing only Chicago addresses overnight” in July and August. In an email sent to Sánchez in August, the school district told her, “Although you are the owner on record of a house in our district boundaries, your license plate recognition shows that is not the place where you reside.”   

But that’s obviously not true. Sanchez says the only reason plate reader data would have shown her car as “staying” in Chicago was because she lent it to a relative during that time period. The school insists this data is enough to overturn the documents she’s provided because… well, it doesn’t really say. It just claims it “relies” on this information gathering to determine residency for students.

All of this can be traced back to Thompson Reuters, which apparently has branched out into the AI-assisted, ALPR-enabled business of denying enrollment to students based on assumptions made by its software.

Here’s what little there is of additional information, as obtained by The Register while reporting on this case:

Thomson Reuters Clear, which more broadly is an AI-assisted records investigation tool, has a page dedicated to its application for school districts. It sells Clear as a tool for residency verification, claiming that it can “automate” such tasks with “enhanced reliability,” and can take care of them “in minutes, not months.” 

One of the particular things the Clear page notes is its ability to access license plate data “and develop pattern of life information” that helps identify whether those who are claiming they’re residents for the sake of getting a kid enrolled in school are lying or not. 

Thomson Reuters does not specify where it gets its license plate reader data and did not respond to questions.

We’ll get to the highlighted sentence in a moment, but let’s just take a beat and consider how creepy and weird this Thomson Reuters promotional pitch is:

The text reads:

Gain deeper insights into mismatched data to support meaningful conversations with families and ensure students are where they need to be. Identify where cars have been seen to establish pattern of life information.

No one expects a law enforcement agency to do this (at least without a warrant or reasonable suspicion), much less a school district. Government agencies shouldn’t have unfettered access to “pattern of life” information just because. It’s not like the people being surveilled here are engaged in criminal activity. They’re just trying to make sure their kids receive an education. And while there will always be people who game the system to get their kids into better schools, that’s hardly justification for subjecting every enrolling student’s family to expansive surveillance-enabled background checks.

And while Thomson Reuters (and the district itself) has refused to comment on the source of its plate reader data, it can safely be assumed that it’s Flock Safety. Flock Safety has never shown any concern about who accesses the data it compiles, much less why they choose to do it. Flock is swiftly becoming the leading provider of ALPR cameras and given its complete lack of internal or external oversight, it’s more than likely the case that its feeding this data to third parties like Thomson Reuters that are willing to pay a premium for data that simply can’t be had elsewhere.

We’re not catching criminals with this tech. Sure, it may happen now and then. But the real value is repeated resale of “pattern of life” data to whoever is willing to purchase it. That’s a massive problem that’s only going to get worse… full stop.

Filed Under: alpr, chicago, license plate readers, surveillance, wtaf

Companies: flock, flock safety, thomson reuters

In the sprawl north of Dallas, where data centres are multiplying and the Texas grid groans under record demand, Energy Vault has placed a new bet on battery storage, and on the idea that the companies powering AI’s insatiable appetite for electricity will need far more of it, far faster.

The California-based energy storage company announced on Monday that it has acquired the McMurtre Battery Energy Storage System (BESS), a 175 MW / 350 MWh project near Dallas, from greenfield developer Belltown Power. The deal advances Energy Vault’s plan, first outlined at its 2025 Investor and Analyst Day, to deploy an initial 1,500 MW of battery storage capacity across the United States and beyond.

Why ERCOT North, and why now

The McMurtre project sits in the ERCOT North market, a region that has become one of the most contested patches of real estate in American energy. Rapid data centre construction near Dallas has driven sustained demand for grid stability and new generation capacity, and power price dynamics in the region remain among the strongest in the country. Energy Vault says it selected the interconnection point specifically for its revenue projections and proximity to that expanding compute infrastructure.

The project already holds an executed Small Generator Interconnection Agreement (SGIA) and full site control, two milestones that significantly de-risk the path to construction. Energy Vault expects to receive Notice to Proceed in the fourth quarter of 2026, with commercial operation targeted for December 2027.

The financial picture

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

According to the company, McMurtre is expected to generate between $15 million and $20 million in average annual revenues over its technical life, translating to an estimated $350 million to $375 million or more in total lifetime revenues. These are forward-looking projections, however, and remain subject to the usual caveats around market conditions, permitting, and execution risk.

Energy Vault intends to contribute the project to its Asset Vault platform, a fully consolidated subsidiary that develops, builds, owns, and operates energy storage assets globally, once it reaches Ready-to-Build status. The company’s $300 million preferred equity investment commitment is designed to support projects like McMurtre as they advance through development and into construction, enabling over $1 billion in total project capital expenditure across the portfolio.

Three asset classes, one thesis

McMurtre is not just a standalone battery project. It fits into a broader strategic architecture that Energy Vault has been assembling around three complementary asset classes: battery energy storage systems, “powered land,” and “powered shells,” the latter referring to modular data centre infrastructure deployed close to energy assets.

That strategy took concrete shape in February 2026, when Energy Vault announced a framework agreement with Crusoe Energy Systems to deploy Crusoe’s Spark modular AI factory units at Energy Vault sites, starting with a 25 MW deployment in Snyder, Texas. The partnership marked the company’s formal entry into AI compute infrastructure and signalled that battery storage, in Energy Vault’s view, is not merely a grid-balancing tool but the foundational layer for a new class of energy-adjacent digital infrastructure.

Robert Piconi, Energy Vault’s chairman and chief executive, framed the acquisition in those terms. The company is building battery assets that enable powered shell deployments, which in turn serve the booming demand for AI compute capacity. McMurtre, he indicated, strengthens that foundation.

A growing portfolio

The acquisition brings Energy Vault’s total owned assets, whether acquired, under construction, or in operation, to 715 MW across all asset classes within its Asset Vault platform. Other projects in the pipeline include the 150 MW / 300 MWh SOSA Energy Center in Texas, the 57 MW / 114 MWh Cross Trails BESS in Texas, an 8.5 MW / 293 MWh resiliency centre in Calistoga, California, and two long-duration storage projects in New South Wales, Australia: the 125 MW / 1.0 GWh Stoney Creek BESS and the 100 MW / 870 MWh Ebor BESS.

The McMurtre system will use Energy Vault’s B-VAULT AC Technology Platform 3, the company’s latest battery product. Globally, the B-VAULT portfolio now exceeds 3 GWh of deployed or contracted systems across Europe, North America, and Australia.

The bigger question

Energy Vault’s wager is ultimately a bet on convergence: that the companies racing to build AI infrastructure will pay a premium for co-located, reliable power, and that vertically integrated storage operators are better positioned than anyone to deliver it. Whether that thesis holds will depend on execution, on ERCOT’s continued growth trajectory, and on whether the AI data centre buildout sustains its current ferocious pace.

For now, the McMurtre deal adds another tile to a mosaic that Energy Vault is assembling across three continents and, increasingly, across the boundary between energy and compute. The grid, it seems, is no longer just about keeping the lights on. It is becoming the scaffolding for something much larger.