OpenAI is gearing up for a coordinated policy and public‑relations push that will launch immediately after Tuesday’s midterm elections. According to Axios, the company’s policy team will roll out a “lame‑duck” campaign timed for the session that opens on Nov. 9, with the centerpiece being a call for mandatory “speed limits” on the most powerful AI systems. The proposal would require developers to impose hard caps on model performance or compute intensity, a safeguard OpenAI says is needed to keep advanced systems under human control.
The timing reflects a calculated response to the likely shift in Congress. Democrats are expected to take control of the House, which analysts say could bring a tougher stance toward the tech sector after two years of a relatively friendly White House. By positioning itself as a proactive regulator, OpenAI hopes to shape the emerging legislative conversation and pre‑empt stricter rules that might otherwise be imposed by a new majority.
Alongside the policy thrust, OpenAI plans to rebrand its data‑center footprint as “AI compute campuses.” The new label is intended to convey a more community‑oriented, infrastructure‑focused image rather than the opaque “data center” moniker that has drawn criticism for its environmental and security implications.
What to watch next: OpenAI’s lobbying efforts in Washington will likely intensify as the lame‑duck session unfolds, with the company expected to meet lawmakers, testify before committees, and release detailed technical proposals for the speed‑limit framework. The reaction of congressional committees, especially those overseeing technology and national security, will indicate whether the industry’s self‑regulation pitch can stave off more aggressive legislation. Keep an eye on statements from the White House’s Office of Science and Technology Policy, which has been a key player in shaping AI policy this year.
Proton’s chief technology officer, Bart Butler, has published an internal “AI manifesto” that declares a seismic shift in how the company’s engineers will work. The document, circulated to the engineering organization this week, argues that large language models (LLMs) have reached a point where they can generate production‑grade code, prompting a re‑examination of the entire software‑development lifecycle. A senior Proton engineer is quoted in the manifesto saying, “No one writes code in {business unit name} anymore,” underscoring the speed of the change.
The manifesto outlines three core themes: treating LLMs as collaborative teammates rather than tools, redesigning workflows to incorporate prompt engineering and model‑output validation, and discarding the notion that “programming” as a solitary activity remains the backbone of engineering work. By framing the transition as “programming is dead, engineering is evolving,” Butler aims to align the team around new practices, tooling, and skill sets.
Why it matters is twofold. First, Proton’s public acknowledgment mirrors a broader industry trend where AI‑driven code generation is moving from experiment to standard practice, as seen in recent AI‑focused acquisitions and funding rounds across the Nordics. Second, the manifesto signals a cultural shift that could reshape hiring, training, and performance metrics within a fast‑growing tech firm.
What to watch next includes Proton’s rollout plan for LLM‑centric development pipelines, any adjustments to its hiring criteria or internal upskilling programs, and whether competitors issue similar guidance. Observers will also be keen to see how the company measures productivity gains and mitigates risks such as model hallucination or security vulnerabilities as it embraces the AI coding era.
Meta has unveiled Muse, an AI‑driven assistant that functions as a commercial surveillance network. According to independent security researcher Karan Joshi and reporting by Wired, TIME and CNET, Muse can tap into users’ private messages, email accounts, financial data and contact lists. Beyond the data of its own users, the system builds detailed relationship maps that include friends, family and other contacts who have never installed the assistant, aggregating these dossiers into anonymised insights that are shared across Meta’s infrastructure.
The revelation matters because it blurs the line between a personal productivity tool and a data‑harvesting engine. By collecting information on non‑users, Muse sidesteps traditional consent frameworks and raises questions about the scope of corporate surveillance under existing privacy laws such as the GDPR and Norway’s Personal Data Act. The ability to stitch together financial, communication and social‑graph data could enable highly granular profiling, potentially feeding targeted advertising, risk‑assessment services or even law‑enforcement requests. For users, the promise of a helpful assistant now carries the risk of exposing their private circles to a corporate data‑broker.
What to watch next includes Meta’s response to the growing scrutiny. The company is likely to face inquiries from data‑protection authorities in the EU and Scandinavia, and may be pressured to revise Muse’s data‑handling policies or to provide clearer opt‑out mechanisms for third‑party contacts. Industry observers will also monitor whether other AI‑assistant providers adopt similar data‑collection models, and how regulators might tighten oversight of AI tools that operate beyond the direct user base. The episode underscores a broader tension between AI convenience and privacy, a theme that has already surfaced in recent coverage of AI agents and their governance.
Saudi Aramco has taken a 49 percent stake in AI startup Humain, injecting roughly $2.6 billion in cash this year and adding its own AI, compute and digital assets to the deal. The remaining 51 percent stays with Saudi Arabia’s Public Investment Fund (PIF), which continues to hold majority control.
The transaction signals Aramco’s push to diversify beyond energy, leveraging its vast financial resources to gain a foothold in the fast‑growing artificial‑intelligence market that underpins gaming, content creation and other digital services. By pairing capital with its own computing infrastructure, Aramco aims to accelerate Humain’s development pipeline and potentially shape the next generation of AI‑driven gaming experiences. For Humain, the partnership offers access to scale‑grade hardware and a deep pool of capital that could speed product rollout and broaden its customer base.
Observers will watch how the combined resources are deployed. Key questions include whether Humain will integrate its technology into existing gaming platforms, how the partnership will affect competition among AI providers, and whether the deal prompts further sovereign‑wealth interest in the sector. The next steps—such as announcements of joint projects, hiring plans or new service offerings—will indicate how quickly the alliance moves from investment to market impact.
Hyperbolic, a startup that aggregates AI‑compute capacity from roughly 30 “neocloud” providers and offers it to enterprises on a rental basis, announced a $30 million Series A‑1 financing round. The round was led by Walden Catalyst, the venture arm of Intel CEO Lip‑Bu Tan, bringing Hyperbolic’s total funding to $50 million since its inception.
The infusion of capital underscores growing investor confidence in a fragmented AI‑infrastructure market. By stitching together a network of smaller cloud operators, Hyperbolic aims to give businesses flexible, on‑demand access to GPU and specialized accelerator resources without the long‑term commitments of traditional cloud contracts. As AI workloads continue to surge across sectors, the ability to source compute from a diversified pool could help firms manage costs, avoid vendor lock‑in and scale more responsively.
Going forward, the company’s next milestones will likely revolve around expanding its neocloud roster, tightening integration with major AI frameworks, and demonstrating cost‑efficiency gains for early customers. Observers will also watch whether the backing of a high‑profile figure like Tan translates into strategic partnerships with larger hardware vendors or accelerates consolidation among niche cloud providers seeking broader market reach.
The White House has told a congressional committee that it is not required to turn over the complete draft of its voluntary AI‑testing framework, invoking national‑security and commercial‑exemption arguments. The response, delivered in a formal letter, declined to produce the key documents that lawmakers had asked to see as part of oversight of the administration’s emerging AI safety regime.
The move matters because the draft framework is intended to guide how companies test advanced models before deployment, a cornerstone of the Trump administration’s broader AI strategy. By withholding the full text, the executive branch limits the ability of legislators and the public to assess whether the safeguards are robust, proportionate, and aligned with democratic values. The cited exemptions also raise questions about how “national security” and proprietary concerns will be balanced against the need for transparency in a sector where rapid innovation can outpace regulation.
The episode follows earlier coverage of the administration’s AI policy architecture, including our profile of White House OSTP Director Michael Kratsios on 1 November, which highlighted the small circle shaping the nation’s AI agenda. The current refusal underscores the tension between rapid policy rollout and congressional demand for insight.
Watch for possible legal challenges to the exemption claims, further FOIA requests, and any subsequent hearings where the administration may be pressed to release a redacted version of the framework. How the White House navigates these pressures will shape the credibility of its voluntary testing approach and could set precedents for future AI governance disclosures.
A new episode of the Equity podcast turned its focus to the growing gap between the promises of artificial intelligence for senior executives and the reality for the wider workforce. The hosts argued that while AI tools can streamline decision‑making, data analysis and even public‑relations for tech CEOs, the same technologies often leave ordinary employees facing heightened expectations, job‑security anxieties and a sense that leadership’s rhetoric is out of touch.
The discussion highlighted recent high‑profile CEO statements that tout AI as a “productivity booster” for the C‑suite, noting how such messaging can appear tone‑deaf when workers are simultaneously dealing with AI‑driven restructuring, performance monitoring and the pressure to up‑skill. By framing AI as a personal convenience for leaders, the narrative risks obscuring broader equity concerns – a theme echoed in our earlier coverage of AI‑induced labour tensions, such as the union‑resistant dynamics explored on 1 November 2026 and Amazon’s mixed‑message memo on AI use.
Why this matters is twofold. First, it underscores a widening perception gap that could fuel employee disengagement and pushback against AI roll‑outs. Second, it raises questions about corporate governance: if AI benefits are concentrated at the top, regulators and investors may demand clearer accountability and more inclusive benefit‑sharing models.
Looking ahead, observers will watch for concrete steps from tech firms to democratise AI gains – for example, transparent up‑skilling programmes, shared profit mechanisms or policy proposals that address AI‑related workplace inequities. The next wave of corporate AI announcements will likely be judged not just on efficiency metrics, but on how well they bridge the divide between CEOs and the employees who ultimately operate the technology.
Heretic 2.0 has been launched, introducing a method for modifying language models without any additional training. The release marks a shift from the traditional approach of fine‑tuning, which typically requires large datasets, compute resources and extended training cycles. Instead, Heretic 2.0 claims to enable users to alter model behaviour on the fly through a set of parameter‑free adjustments.
The significance of a training‑free modification tool lies in its potential to democratise model customization. Developers and organisations that lack the infrastructure for full‑scale retraining could now tailor outputs to specific domains, languages or safety constraints more quickly and cheaply. At the same time, the ease of alteration raises questions about governance: if model behaviour can be changed without traceable training runs, monitoring for malicious or biased modifications may become more challenging for regulators and platform operators.
What to watch next includes the community’s technical assessment of the method’s robustness and limits. Early adopters are likely to publish benchmarks comparing Heretic 2.0’s modifications against conventional fine‑tuning in terms of performance, safety and resource consumption. Parallel discussions are expected in policy circles about how to audit and certify models that have been altered post‑deployment. Finally, the developers have hinted at further updates that could expand the range of supported model families, making the upcoming months a key period for both practitioners and watchdogs tracking the evolution of training‑free AI tooling.
