Stay Ahead, Stay ONMINE

OpenAI’s red teaming innovations define new essentials for security leaders in the AI era

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More OpenAI has taken a more aggressive approach to red teaming than its AI competitors, demonstrating its security teams’ advanced capabilities in two areas: multi-step reinforcement and external red teaming. OpenAI recently released two papers that set […]

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More


OpenAI has taken a more aggressive approach to red teaming than its AI competitors, demonstrating its security teams’ advanced capabilities in two areas: multi-step reinforcement and external red teaming. OpenAI recently released two papers that set a new competitive standard for improving the quality, reliability and safety of AI models in these two techniques and more.

The first paper, “OpenAI’s Approach to External Red Teaming for AI Models and Systems,” reports that specialized teams outside the company have proven effective in uncovering vulnerabilities that might otherwise have made it into a released model because in-house testing techniques may have missed them.

In the second paper, “Diverse and Effective Red Teaming with Auto-Generated Rewards and Multi-Step Reinforcement Learning,” OpenAI introduces an automated framework that relies on iterative reinforcement learning to generate a broad spectrum of novel, wide-ranging attacks.

Going all-in on red teaming pays practical, competitive dividends

It’s encouraging to see competitive intensity in red teaming growing among AI companies. When Anthropic released its AI red team guidelines in June of last year, it joined AI providers including Google, Microsoft, Nvidia, OpenAI, and even the U.S.’s National Institute of Standards and Technology (NIST), which all had released red teaming frameworks.

Investing heavily in red teaming yields tangible benefits for security leaders in any organization. OpenAI’s paper on external red teaming provides a detailed analysis of how the company strives to create specialized external teams that include cybersecurity and subject matter experts. The goal is to see if knowledgeable external teams can defeat models’ security perimeters and find gaps in their security, biases and controls that prompt-based testing couldn’t find.

What makes OpenAI’s recent papers noteworthy is how well they define using human-in-the-middle design to combine human expertise and contextual intelligence on one side with AI-based techniques on the other.

“When automated red teaming is complemented by targeted human insight, the resulting defense strategy becomes significantly more resilient,” writes OpenAI in the first paper (Ahmad et al., 2024).

The company’s premise is that using external testers to identify the most high-impact real-world scenarios, while also evaluating AI outputs, leads to continuous model improvements. OpenAI contends that combining these methods delivers a multi-layered defense for their models that identify potential vulnerabilities quickly. Capturing and improving models with the human contextual intelligence made possible by a human-in-the-middle design is proving essential for red-teaming AI models.

Why red teaming is the strategic backbone of AI security

Red teaming has emerged as the preferred method for iteratively testing AI models. This kind of testing simulates a variety of lethal and unpredictable attacks and aims to identify their most potent and weakest points. Generative AI (gen AI) models are difficult to test through automated means alone, as they mimic human-generated content at scale. The practices described in OpenAI’s two papers seek to close the gaps automated testing alone leaves, by measuring and verifying a model’s claims of safety and security.

In the first paper (“OpenAI’s Approach to External Red Teaming”) OpenAI explains that red teaming is “a structured testing effort to find flaws and vulnerabilities in an AI system, often in a controlled environment and collaboration with developers” (Ahmad et al., 2024). Committed to leading the industry in red teaming, the company had over 100 external red teamers assigned to work across a broad base of adversarial scenarios during the pre-launch vetting of GPT-4 prior to launch.

Research firm Gartner reinforces the value of red teaming in its forecast, predicting that IT spending on gen AI will soar from $5 billion in 2024 to $39 billion by 2028. Gartner notes that the rapid adoption of gen AI and the proliferation of LLMs is significantly expanding these models’ attack surfaces, making red teaming essential in any release cycle.

Practical insights for security leaders

Even though security leaders have been quick to see the value of red teaming, few are following through by making a commitment to get it done. A recent Gartner survey finds that while 73% of organizations recognize the importance of dedicated red teams, only 28% actually maintain them. To close this gap, a simplified framework is needed that can be applied at scale to any new model, app, or platform’s red teaming needs.

In its paper on external red teaming OpenAI defines four key steps for using a human-in-the-middle design to make the most of human insights:

  • Defining testing scope and teams: Drawing on subject matter experts and specialists across key areas of cybersecurity, regional politics, and natural sciences, OpenAI targets risks that include voice mimicry and bias. The ability to recruit cross-functional experts is, therefore, crucial. (To gain an appreciation for how committed OpenAI is to this methodology and its implications for stopping deepfakes, please see our article “GPT-4: OpenAI’s shield against $40B deepfake threat to enterprises.”)
  • Selecting model versions for testing, then iterating them across diverse teams: Both of OpenAI’s papers emphasize that cycling red teams and models using an iterative approach delivers the most insightful results. Allowing each red team to cycle through all models is conducive to greater team learning of what is and isn’t working.
  • Clear documentation and guidance: Consistency in testing requires well-documented APIs, standardized report formats, and explicit feedback loops. These are essential elements for successful red teaming.
  • Making sure insights translate into practical and long-lasting mitigations: Once red teams log vulnerabilities, they drive targeted updates to models, policies and operational plans — ensuring security strategies evolve in lockstep with emerging threats.

Scaling adversarial testing with GPT-4T: The next frontier in red teaming

AI companies’ red teaming methodologies are demonstrating that while human expertise is resource-intensive, it remains crucial for in-depth testing of AI models.

In OpenAI’s second paper, “Diverse and Effective Red Teaming with Auto-Generated Rewards and Multi-Step Reinforcement Learning” (Beutel et al., 2024), OpenAI addresses the challenge of scaling adversarial testing using an automated, multi-pronged approach that combines human insights with AI-generated attack strategies.

The core of this methodology is GPT-4T, a specialized variant of the GPT-4 model engineered to produce a wide range of adversarial scenarios.

Here’s how each component of the methodology contributes to a stronger adversarial testing framework:

  • Goal diversification. OpenAI describes how it is using GPT-4T to create a broad spectrum of scenarios, starting with initially benign-seeming prompts and progressing to more sophisticated phishing campaigns. Goal diversification focuses on anticipating and exploring the widest possible range of potential exploits. By using GPT-4T’s capacity for diverse language generation, OpenAI contends that red teams avoid tunnel vision and stay focused on probing for vulnerabilities that manual-only methods miss.
  • Reinforcement learning (RL). A multi-step RL framework rewards the discovery of new and previously unseen vulnerabilities. The purpose is to train the automated red team by improving each iteration. This enables security leaders to refocus on genuine risks rather than sifting through volumes of low-impact alerts. It aligns with Gartner’s projection of a 30% drop in false positives attributable to gen AI in application security testing by 2027. OpenAI writes, “Our multi-step RL approach systematically rewards the discovery of newly identified vulnerabilities, driving continuous improvement in adversarial testing.”
  • Auto-generated rewards: OpenAI defines this as a system that tracks and updates scores for partial successes by red teams, assigning incremental rewards for identifying each unprotected weak area of a model.

Securing the future of AI: Key takeaways for security leaders

OpenAI’s recent papers show why a structured, iterative process that combines internal and external testing delivers the insights needed to keep improving models’ accuracy, safety, security and quality.

Security leaders’ key takeaways from these papers should include: 

Go all-in and adopt a multi-pronged approach to red teaming. The papers emphasize the value of combining external, human-led teams with real-time simulations of AI attacks generated randomly, as they reflect how chaotic intrusion attempts can be. OpenAI contends that while humans excel at spotting context-specific gaps, including biases, automated systems identify weaknesses that emerge only under stress testing and repeated sophisticated attacks.

Test early and continuously throughout model dev cycles. The white papers make a compelling argument against waiting for production-ready models and instead beginning testing with early-stage versions. The goal is to find emerging risks and retest later to make sure the gaps in models were closed before launch.

Whenever possible, streamline documentation and feedback with real-time feedback loops. Standardized reporting and well-documented APIs, along with explicit feedback loops, help convert red team findings into actionable, trackable mitigations. OpenAI emphasizes the need to get this process in place before beginning red teaming, to accelerate fixes and remediation of problem areas.

Using real-time reinforcement learning is critically important, as is the future of AI red teaming. OpenAI makes the case for automating frameworks that reward discoveries of new attack vectors as a core part of the real-time feedback loops. The goal of RL is to create a continuous loop of improvement. 

Don’t settle for anything less than actionable insights from the red team process. It’s essential to treat every red team discovery or finding as a catalyst for updating security strategies, improving incident response plans, and revamping guidelines as required.

Budget for the added expense of enlisting external expertise for red teams. A central premise of OpenAI’s approach to red teaming is to actively recruit outside specialists who have informed perspectives and knowledge of advanced threats. Areas of expertise valuable to AI-model red teams include deepfake technology, social engineering, identity theft, synthetic identity creation, and voice-based fraud. “Involving external specialists often surfaces hidden attack paths, including sophisticated social engineering and deepfake threats.” (Ahmad et al., 2024)

Papers:

Beutel, A., Xiao, K., Heidecke, J., & Weng, L. (2024). “Diverse and Effective Red Teaming with Auto-Generated Rewards and Multi-Step Reinforcement Learning.” OpenAI.

Ahmad, L., Agarwal, S., Lampe, M., & Mishkin, P. (2024). “OpenAI’s Approach to External Red Teaming for AI Models and Systems.” OpenAI.

Shape
Shape
Stay Ahead

Explore More Insights

Stay ahead with more perspectives on cutting-edge power, infrastructure, energy,  bitcoin and AI solutions. Explore these articles to uncover strategies and insights shaping the future of industries.

Shape

Ubuntu namespace vulnerability should be addressed quickly: Expert

Thus, “there is little impact of not ‘patching’ the vulnerability,” he said. “Organizations using centralized configuration tools like Ansible may deploy these changes with regularly scheduled maintenance or reboot windows.”  Features supposed to improve security Ironically, last October Ubuntu introduced AppArmor-based features to improve security by reducing the attack surface

Read More »

Dallas Fed survey shows capex plans holding up despite uncertainty

And yet: Despite those generally upbeat readings, respondents’ comments to Fed analysts were very focused on the uncertainty that the Trump administration’s energy and trade policies have created over the past 10 weeks. That instability, they said, as well as the downward trend in oil prices has many taking a more cautious view of the future even if those attitudes aren’t yet showing up in capex intentions. “Planning for new development is extremely difficult right now due to the uncertainty around steel-based products,” one executive wrote. “Oil prices feel incredibly unstable, and it’s hard to gauge whether prices will be in the $50s per barrel or $70s per barrel. Combined, our ability to plan operations for any meaningful amount of time in the future has been severely diminished.” WTI thresholds for opex, investment Fed researchers this quarter also asked E&P companies some one-time questions about where they need West Texas Intermediate prices to be to cover the operating costs of existing wells as well as to profitably drill a new well. While the central bank asked the same question early last year, the timing of this survey was especially topical given recent comments by US Secretary of Energy to the Financial Times that the industry could still grow production profitably even if WTI fell to $50/bbl.  Not so fast, respondents to the Dallas Fed poll said: On the opex question, their responses averaged $41/bbl (up from $39 a year ago) and ranged from $26 in the Eagle Ford to $45 in parts of the Permian outside the Delaware and Midland basins. But when it comes to justifying investments in new wells, respondents said they need at least $61 (in the Midland) to $70 in the ‘other’ parts of the Permian. The average price they need to profitably drill a new

Read More »

Equinor, Shell, TotalEnergies take FID on Northern Lights CCS expansion

Equinor, Shell, and TotalEnergies have made a final investment decision (FID) to progress phase two of the Northern Lights carbon capture development.  The investment by the Northern Lights JV owners (Equinor, Shell, TotalEnergies) is about $714 million. The expansion project received €131 million from the Connecting Europe Facility (CEF) in June 2024. The news comes following a signed commercial agreement with Stockholm Exergi to transport and store 900,000 tonnes/year (tpy) of biogenic CO2 for 15 years. The Northern Lights project comprises transportation, receipt, and permanent storage of CO2 in a reservoir in the northern North Sea. Northern Lights phases The first phase includes capacity to transport, inject, and store up to 1.5 million tpy of CO2. Once the CO2 is captured onshore, it will be transported by ship to the receiving terminal in Øygarden, pumped via pipeline to a subsea structure at the seabed and injected into a geological formation some 2,500 m below the seabed in the North Sea for permanent storage. Phase one operations are planned for this summer, with CO2 from Heidelberg Materials’ cement factory in Brevik expected to arrive at the receiving terminal near Kollsnes on Norway’s west coast. Additionally, Northern Lights will store CO2 from the Hafslund Celsio waste-to-energy plant in Oslo, as part of the Longship project (OGJ Online, Oct. 9, 2024).

Read More »

Infinity Natural Resources plans capex ramp, greater gas emphasis

Leaders of Infinity Natural Resources Inc., Morgantown, W.Va., plan to ramp up capital spending this year across their Appalachian holdings and tilt their development work more toward natural gas. Infinity, which went public earlier this year, spent $166 million last year on drilling and completion work in the Appalachian basin—where it produces oil from the Ohio Utica basin and gas from holdings in the Utica and Marcellus regions—as well as $5.5 million on its midstream assets and $108 million on land. This year, president and chief executive officer Zack Arnold told analysts on a conference call that his team will have “limited need” to add to its land holdings. But Infinity’s leaders are forecasting that their drilling and completions spending will grow this year to $240-280 million as the operator looks to capitalize on improving gas market fundamentals. Spending on midstream assets is expected to grow to $9-12 million. Infinity’s portfolio comprises about 60,000 acres in Pennsylvania that at end-2024 sported 179 undeveloped locations as well as roughly 63,000 net acres in the Utica’s volatile oil window, where it had 154 undeveloped locations. The operator’s 2024 production averaged 24,100 boe/d (27% oil, 53% gas, 20% natural gas liquids), which was an increase of 28% from the year before thanks in large part to the addition of 12 net wells in the Utica. Arnold and his team are planning to grow production to 32,000-35,000 boe/d this year. The company expects to run one rig for the year with a second rig added to initially develop a four-well pad in the Marcellus. “Our 2025 plan highlights a transition towards a greater balance between natural gas and oil-weighted wells,” Arnold said on the call after noting the oil wells turned in line last year as well as the deferral of completion work on

Read More »

Equinor begins production at Johan Castberg

Equinor Energy AS began production at Johan Castberg oil field in the Barents Sea on Mar. 31, 2025. The field is expected to produce for 30 years. Twelve of the 30 total wells are ready for production, sufficient to bring the field up to expected plateau production in second-quarter 2025, the operator said in a release Mar. 31. Drilling operations are expected to continue towards late 2026. Johan Castberg lies 100 km north of Snøhvit field in Blocks 7219/9 and 7220/4,5,7 in 360-390 m of water. The field consists of Skrugard, Havis, and Drivis discoveries made between 2011 and 2014. It is the second oil field in the Barents Sea and Norway’s northernmost field. Field development is based on a production vessel tied back to an extensive subsea field with a total of 30 wells distributed between 10 well templates and two satellite structures (OGJ Online, Dec. 10, 2024). The Equinor-operated Johan Castberg FPSO has a gross capacity of 220,000 b/d of oil. Its design storage capacity is 1.1 million bbl of oil. The field holds gross recoverable volumes of 450-650 million bbl of oil.   The Johan Castberg area holds upside as several new discoveries made in recent years are already being matured into projects, including Johan Castberg Cluster 1, said partner Vår Energi ASA. Cluster 2 is progressing through near field exploration, and an extensive infill drilling program is being planned, the company said. A total of 250-550 million bbl of oil of additional gross unrisked recoverable resources have been identified in the area. Equinor is operator (46.3%) with partners Vår Energi ASA (30%) and Petoro AS (23.7%).

Read More »

Market Focus: Insights from Oil & Gas Journal’s latest capital spending survey

@import url(‘/fonts/fira_sans.css’); a { color: #134e85; } .ebm-page__main h1, .ebm-page__main h2, .ebm-page__main h3, .ebm-page__main h4, .ebm-page__main h5, .ebm-page__main h6 { font-family: “Fira Sans”, Arial, sans-serif; } body { letter-spacing: 0.025em; font-family: “Fira Sans”, Arial, sans-serif; } button, .ebm-button-wrapper { font-family: “Fira Sans”, Arial, sans-serif; } .label-style { text-transform: uppercase; color: var(–color-grey); font-weight: 600; font-size: 0.75rem; } .caption-style { font-size: 0.75rem; opacity: .6; } #onetrust-pc-sdk [id*=btn-handler], #onetrust-pc-sdk [class*=btn-handler] { background-color: #212529 !important; border-color: #212529 !important; } #onetrust-policy a, #onetrust-pc-sdk a, #ot-pc-content a { color: #212529 !important; } #onetrust-consent-sdk #onetrust-pc-sdk .ot-active-menu { border-color: #212529 !important; } #onetrust-consent-sdk #onetrust-accept-btn-handler, #onetrust-banner-sdk #onetrust-reject-all-handler, #onetrust-consent-sdk #onetrust-pc-btn-handler.cookie-setting-link { background-color: #212529 !important; border-color: #212529 !important; } #onetrust-consent-sdk .onetrust-pc-btn-handler { color: #212529 !important; border-color: #212529 !important; background-color: undefined !important; } <!–> In this latest Market Focus episode of the Oil & Gas Journal ReEnterprised podcast, Conglin Xu, Managing Editor, Economics, dives into insights from the latest Oil & Gas Journal capital spending report.  According to the OGJ annual capital spending survey, the combined capex of six major oil companies—ExxonMobil, Chevron, Shell, BP, Equinor, and TotalEnergies—is projected to be US$108-112 billion in 2025. This marks a decrease from $113.7 billion in 2024 and $114.7 billion in 2023 and remains significantly lower than the pre-pandemic level of $123 billion in 2019. Notably, majors are scaling back on earlier aggressive investments in renewables.  From shifting strategies among oil majors to merger and acquisition activity in the shale sector to new developments in the refining sector and the Canadian oil industry, there’s a lot to unpack.  ]–>          FXQuardro/Shutterstock.com <!–> ]–> <!–> OGJ Premium Members can read the full Capital Spending Update from the Mar/Apr 2025 issue.  ]–>

Read More »

TGNR adds East Texas gas assets in $525-million deal with Chevron

The deal adds over 250 gross locations to TGNR’s existing Haynesville inventory (assuming four wells per section), extending inventory life beyond 20 years at the current development pace, not counting the Bossier and Cotton Valley plays which are commercial at current prices, the company said in a release Mar. 31. According to its website, Chevron holds about 72,000 net acres (283 sq km) in the Haynesville shale in East Texas as of Dec. 31, 2024.  TGNR said the deal’s acreage is “relatively undrilled and held by shallower production, allowing parent-child effects between wells to be mitigated.” The company expects to realize synergies of over $170 million during the assets’ development.  Chevron, in a separate release Mar. 31, said the deal is expected to generate over $1.2 billion in value to the company at current Henry Hub prices through the multi-year capital carry, retained working interest, and overriding royalty interest. For Chevron, the deal with TGNR supports its plans to divest $10-15 billion of assets by 2028. TGNR is focused on the Ark-La-Tex region of East Texas and Northern Louisiana. It is owned by TG East Texas Resources LLC, a wholly owned subsidiary of Tokyo Gas America, and CCI US Asset Holdings LLC, a wholly owned subsidiary of Castleton Commodities International LLC.  The purchase price is comprised of $75 million paid in cash and $450 million as a capital carry to fund Haynesville development. 

Read More »

European cloud group invests to create what it dubs “Trump-proof cloud services”

But analysts have questioned whether the Microsoft move truly addresses those European business concerns. Phil Brunkard, executive counselor at Info-Tech Research Group UK, said, commenting on last month’s announcement of the EU Data Boundary for the Microsoft Cloud,  “Microsoft says that customer data will remain stored and processed in the EU and EFTA, but doesn’t guarantee true data sovereignty.” And European companies are now rethinking what data sovereignty means to them. They are moving beyond having it refer to where the data sits to focusing on which vendors control it, and who controls them. Responding to the new Euro cloud plan, another analyst, IDC VP Dave McCarthy, saw the effort as “signaling a growing European push for data control and independence.” “US providers could face tougher competition from EU companies that leverage this tech to offer sovereignty-friendly alternatives. Although €1 million isn’t a game-changer on its own, it’s a clear sign Europe wants to build its own cloud ecosystem—potentially at the expense of US market share,” McCarthy said. “For US providers, this could mean investing in more EU-based data centers or reconfiguring systems to ensure European customers’ data stays within the region. This isn’t just a compliance checkbox. It’s a shift that could hike operational costs and complexity, especially for companies used to running centralized setups.” Adding to the potential bad news for US hyperscalers, McCarthy said that there was little reason to believe that this trend would be limited to Europe. “If Europe pulls this off, other regions might take note and push for similar sovereignty rules. US providers could find themselves adapting to a patchwork of regulations worldwide, forcing a rethink of their global strategies,” McCarthy said. “This isn’t just a European headache, it’s a preview of what could become a broader challenge.”

Read More »

Talent gap complicates cost-conscious cloud planning

The top strategy so far is what one enterprise calls the “Cloud Team.” You assemble all your people with cloud skills, and your own best software architect, and have the team examine current and proposed cloud applications, looking for a high-level approach that meets business goals. In this process, the team tries to avoid implementation specifics, focusing instead on the notion that a hybrid application has an agile cloud side and a governance-and-sovereignty data center side, and what has to be done is push functionality into the right place. The Cloud Team supporters say that an experienced application architect can deal with the cloud in abstract, without detailed knowledge of cloud tools and costs. For example, the architect can assess the value of using an event-driven versus transactional model without fixating on how either could be done. The idea is to first come up with approaches. Then, developers could work with cloud providers to map each approach to an implementation, and assess the costs, benefits, and risks. Ok, I lied about this being the top strategy—sort of, at least. It’s the only strategy that’s making much sense. The enterprises all start their cloud-reassessment journey on a different tack, but they agree it doesn’t work. The knee-jerk approach to cloud costs is to attack the implementation, not the design. What cloud features did you pick? Could you find ones that cost less? Could you perhaps shed all the special features and just host containers or VMs with no web services at all? Enterprises who try this, meaning almost all of them, report that they save less than 15% on cloud costs, a rate of savings that means roughly a five-year payback on the costs of making the application changes…if they can make them at all. Enterprises used to build all of

Read More »

Lightmatter launches photonic chips to eliminate GPU idle time in AI data centers

“Silicon photonics can transform HPC, data centers, and networking by providing greater scalability, better energy efficiency, and seamless integration with existing semiconductor manufacturing and packaging technologies,” Jagadeesan added. “Lightmatter’s recent announcement of the Passage L200 co-packaged optics and M1000 reference platform demonstrates an important step toward addressing the interconnect bandwidth and latency between accelerators in AI data centers.” The market timing appears strategic, as enterprises worldwide face increasing computational demands from AI workloads while simultaneously confronting the physical limitations of traditional semiconductor scaling. Silicon photonics offers a potential path forward as conventional approaches reach their limits. Practical applications For enterprise IT leaders, Lightmatter’s technology could impact several key areas of infrastructure planning. AI development teams could see significantly reduced training times for complex models, enabling faster iteration and deployment of AI solutions. Real-time AI applications could benefit from lower latency between processing units, improving responsiveness for time-sensitive operations. Data centers could potentially achieve higher computational density with fewer networking bottlenecks, allowing more efficient use of physical space and resources. Infrastructure costs might be optimized by more efficient utilization of expensive GPU resources, as processors spend less time waiting for data and more time computing. These benefits would be particularly valuable for financial services, healthcare, research institutions, and technology companies working with large-scale AI deployments. Organizations that rely on real-time analysis of large datasets or require rapid training and deployment of complex AI models stand to gain the most from the technology. “Silicon photonics will be a key technology for interconnects across accelerators, racks, and data center fabrics,” Jagadeesan pointed out. “Chiplets and advanced packaging will coexist and dominate intra-package communication. The key aspect is integration, that is companies who have the potential to combine photonics, chiplets, and packaging in a more efficient way will gain competitive advantage.”

Read More »

Silicon Motion rolls SSD kit to bolster AI workload performance

The kit utilizes the PCIe Dual Ported enterprise-grade SM8366 controller with support for PCIe Gen 5 x4 NVMe 2.0 and OCP 2.5 data center specifications. The 128TB SSD RDK also supports NVMe 2.0 Flexible Data Placement (FDP), a feature that allows advanced data management and improved SSD write efficiency and endurance. “Silicon Motion’s MonTitan SSD RDK offers a comprehensive solution for our customers, enabling them to rapidly develop and deploy enterprise-class SSDs tailored for AI data center and edge server applications.” said Alex Chou, senior vice president of the enterprise storage & display interface solution business at Silicon Motion. Silicon Motion doesn’t make drives, rather it makes reference design kits in different form factors that its customers use to build their own product. Its kits come in E1.S, E3.S, and U.2 form factors. The E1.S and U.2 forms mirror the M.2, which looks like a stick of gum and installs on the motherboard. There are PCI Express enclosures that hold four to six of those drives and plug into one card slot and appear to the system as a single drive.

Read More »

Executive Roundtable: Cooling Imperatives for Managing High-Density AI Workloads

Michael Lahoud, Stream Data Centers: For the past two years, Stream Data Centers has been developing a modular, configurable air and liquid cooling system that can handle the highest densities in both mediums. Based on our collaboration with customers, we see a future that still requires both cooling mediums, but with the flexibility to deploy either type as the IT stack destined for that space demands. With this necessity as a backdrop, we saw a need to develop a scalable mix-and-match front-end thermal solution that gives us the ability to late bind the equipment we need to meet our customers’ changing cooling needs. It’s well understood that liquid far outperforms air in its ability to transport heat, but further to this, with the right IT configuration, cooling fluid temperatures can also be raised, and this affords operators the ability to use economization for a greater number of hours a year. These key properties can help reduce the energy needed for the mechanical part of a data center’s operations substantially.  It should also be noted that as servers are redesigned for liquid cooling and the onboard server fans get removed or reduced in quantity, more of the critical power delivered to the server is being used for compute. This means that liquid cooling also drives an improvement in overall compute productivity despite not being noted in facility PUE metrics.  Counter to air cooling, liquid cooling certainly has some added management challenges related to fluid cleanliness, concurrent maintainability and resiliency/redundancy, but once those are accounted for, the clusters become stable, efficient and more sustainable with improved overall productivity.

Read More »

Airtel connects India with 100Tbps submarine cable

“Businesses are becoming increasingly global and digital-first, with industries such as financial services, data centers, and social media platforms relying heavily on real-time, uninterrupted data flow,” Sinha added. The 2Africa Pearls submarine cable system spans 45,000 kilometers, involving a consortium of global telecommunications leaders including Bayobab, China Mobile International, Meta, Orange, Telecom Egypt, Vodafone Group, and WIOCC. Alcatel Submarine Networks is responsible for the cable’s manufacturing and installation, the statement added. This cable system is part of a broader global effort to enhance international digital connectivity. Unlike traditional telecommunications infrastructure, the 2Africa Pearls project represents a collaborative approach to solving complex global communication challenges. “The 100 Tbps capacity of the 2Africa Pearls cable significantly surpasses most existing submarine cable systems, positioning India as a key hub for high-speed connectivity between Africa, Europe, and Asia,” said Prabhu Ram, VP for Industry Research Group at CyberMedia Research. According to Sinha, Airtel’s infrastructure now spans “over 400,000 route kilometers across 34+ cables, connecting 50 countries across five continents. This expansive infrastructure ensures businesses and individuals stay seamlessly connected, wherever they are.” Gogia further emphasizes the broader implications, noting, “What also stands out is the partnership behind this — Airtel working with Meta and center3 signals a broader shift. India is no longer just a consumer of global connectivity. We’re finally shaping the routes, not just using them.”

Read More »

Microsoft will invest $80B in AI data centers in fiscal 2025

And Microsoft isn’t the only one that is ramping up its investments into AI-enabled data centers. Rival cloud service providers are all investing in either upgrading or opening new data centers to capture a larger chunk of business from developers and users of large language models (LLMs).  In a report published in October 2024, Bloomberg Intelligence estimated that demand for generative AI would push Microsoft, AWS, Google, Oracle, Meta, and Apple would between them devote $200 billion to capex in 2025, up from $110 billion in 2023. Microsoft is one of the biggest spenders, followed closely by Google and AWS, Bloomberg Intelligence said. Its estimate of Microsoft’s capital spending on AI, at $62.4 billion for calendar 2025, is lower than Smith’s claim that the company will invest $80 billion in the fiscal year to June 30, 2025. Both figures, though, are way higher than Microsoft’s 2020 capital expenditure of “just” $17.6 billion. The majority of the increased spending is tied to cloud services and the expansion of AI infrastructure needed to provide compute capacity for OpenAI workloads. Separately, last October Amazon CEO Andy Jassy said his company planned total capex spend of $75 billion in 2024 and even more in 2025, with much of it going to AWS, its cloud computing division.

Read More »

John Deere unveils more autonomous farm machines to address skill labor shortage

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Self-driving tractors might be the path to self-driving cars. John Deere has revealed a new line of autonomous machines and tech across agriculture, construction and commercial landscaping. The Moline, Illinois-based John Deere has been in business for 187 years, yet it’s been a regular as a non-tech company showing off technology at the big tech trade show in Las Vegas and is back at CES 2025 with more autonomous tractors and other vehicles. This is not something we usually cover, but John Deere has a lot of data that is interesting in the big picture of tech. The message from the company is that there aren’t enough skilled farm laborers to do the work that its customers need. It’s been a challenge for most of the last two decades, said Jahmy Hindman, CTO at John Deere, in a briefing. Much of the tech will come this fall and after that. He noted that the average farmer in the U.S. is over 58 and works 12 to 18 hours a day to grow food for us. And he said the American Farm Bureau Federation estimates there are roughly 2.4 million farm jobs that need to be filled annually; and the agricultural work force continues to shrink. (This is my hint to the anti-immigration crowd). John Deere’s autonomous 9RX Tractor. Farmers can oversee it using an app. While each of these industries experiences their own set of challenges, a commonality across all is skilled labor availability. In construction, about 80% percent of contractors struggle to find skilled labor. And in commercial landscaping, 86% of landscaping business owners can’t find labor to fill open positions, he said. “They have to figure out how to do

Read More »

2025 playbook for enterprise AI success, from agents to evals

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More 2025 is poised to be a pivotal year for enterprise AI. The past year has seen rapid innovation, and this year will see the same. This has made it more critical than ever to revisit your AI strategy to stay competitive and create value for your customers. From scaling AI agents to optimizing costs, here are the five critical areas enterprises should prioritize for their AI strategy this year. 1. Agents: the next generation of automation AI agents are no longer theoretical. In 2025, they’re indispensable tools for enterprises looking to streamline operations and enhance customer interactions. Unlike traditional software, agents powered by large language models (LLMs) can make nuanced decisions, navigate complex multi-step tasks, and integrate seamlessly with tools and APIs. At the start of 2024, agents were not ready for prime time, making frustrating mistakes like hallucinating URLs. They started getting better as frontier large language models themselves improved. “Let me put it this way,” said Sam Witteveen, cofounder of Red Dragon, a company that develops agents for companies, and that recently reviewed the 48 agents it built last year. “Interestingly, the ones that we built at the start of the year, a lot of those worked way better at the end of the year just because the models got better.” Witteveen shared this in the video podcast we filmed to discuss these five big trends in detail. Models are getting better and hallucinating less, and they’re also being trained to do agentic tasks. Another feature that the model providers are researching is a way to use the LLM as a judge, and as models get cheaper (something we’ll cover below), companies can use three or more models to

Read More »

OpenAI’s red teaming innovations define new essentials for security leaders in the AI era

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More OpenAI has taken a more aggressive approach to red teaming than its AI competitors, demonstrating its security teams’ advanced capabilities in two areas: multi-step reinforcement and external red teaming. OpenAI recently released two papers that set a new competitive standard for improving the quality, reliability and safety of AI models in these two techniques and more. The first paper, “OpenAI’s Approach to External Red Teaming for AI Models and Systems,” reports that specialized teams outside the company have proven effective in uncovering vulnerabilities that might otherwise have made it into a released model because in-house testing techniques may have missed them. In the second paper, “Diverse and Effective Red Teaming with Auto-Generated Rewards and Multi-Step Reinforcement Learning,” OpenAI introduces an automated framework that relies on iterative reinforcement learning to generate a broad spectrum of novel, wide-ranging attacks. Going all-in on red teaming pays practical, competitive dividends It’s encouraging to see competitive intensity in red teaming growing among AI companies. When Anthropic released its AI red team guidelines in June of last year, it joined AI providers including Google, Microsoft, Nvidia, OpenAI, and even the U.S.’s National Institute of Standards and Technology (NIST), which all had released red teaming frameworks. Investing heavily in red teaming yields tangible benefits for security leaders in any organization. OpenAI’s paper on external red teaming provides a detailed analysis of how the company strives to create specialized external teams that include cybersecurity and subject matter experts. The goal is to see if knowledgeable external teams can defeat models’ security perimeters and find gaps in their security, biases and controls that prompt-based testing couldn’t find. What makes OpenAI’s recent papers noteworthy is how well they define using human-in-the-middle

Read More »