12 stories · 3 sources · 1 critical · ~13 min read

Coverage: Last 24 hours

Today’s Highlights

A human attacker weaponized the Marimo remote code execution (RCE) vulnerability (CVE-2026-39987) and breached core infrastructure in mere seconds, emphasizing the risks from unpatched systems and the necessity of robust segmentation and SSH control. At the same time, the absence of US government action on AI regulation and mounting debate among industry leaders are increasing uncertainty and compliance ambiguity for organizations deploying AI. Today’s coverage captures these escalating risks and shifting regulatory climates, offering critical context for defenders on both technical and governance fronts.

Table of Contents

1. Human Attacker Exploits Marimo RCE, Reaches SSH Bastion in Eight Seconds
2. The US government is failing Americans on AI | Shakeel Hashim
3. ‘Pacing’ won’t eliminate the risk of AI doom. Here’s what could | David Krueger
4. Decoding America: Why won’t Donald Trump regulate AI? – podcast
5. Wednesday briefing: Why tech companies might be only too happy for us to believe AI will ‘kill us all’
6. The Spin | Robots bowling the perfect doosra are some way off but AI is already reshaping cricket
7. I used to think AI would kill us all, until the techbro CEOs said AI will kill us all and now I’m not so sure | First Dog on the Moon

Critical   High   Medium   Low

Top Stories

Human Attacker Exploits Marimo RCE, Reaches SSH Bastion in Eight Seconds

Source: The Hacker News | Published: Sep 15 | Risk: CRITICAL | Impacted: Marimo appliance administrators, SSH bastion hosts, Organizations exposing Marimo to the internet | Topics: Exploit / Vulnerability

What happened: A human attacker exploited the Marimo RCE vulnerability (CVE-2026-39987) to gain SSH access to a bastion host in eight seconds, using a custom Python toolkit without AI assistance.

Why it matters: A newly disclosed remote code execution flaw in Marimo was weaponized instantly to gain deep access to core infrastructure, demonstrating how little time organizations have to respond before attackers pivot through exposed systems.

How it works: Marimo is an enterprise platform with an identified remote code execution vulnerability (CVE-2026-39987), allowing attackers to run arbitrary code on the device remotely. Exploitation enables attackers to pivot from Marimo into internal SSH bastions if network paths and credential reuse exist.

Practitioner Perspective

Enterprises running Marimo need to recognize proof-of-concept code for CVE-2026-39987 is now in active use by skilled human adversaries, not just automated bots. The attack shows how quickly an RCE can be chained directly to compromise SSH-protected assets if segmentation and credential hygiene are weak. Security teams should not assume firewall or bastion isolation alone provides meaningful delay against post-exploitation movement. Map all internet-facing Marimo instances and validate that any SSH service reachable from these points is protected by MFA and hardened policies. Your incident response plan for a Marimo breach should include rapid SSH key rotation and cross-segment audit for further lateral movement.

Recommended Actions

• Immediately patch all Marimo systems for CVE-2026-39987 or apply vendor-recommended mitigations
• Hunt for evidence of CVE-2026-39987 exploitation and subsequent SSH access attempts in SIEM/EDR logs

The US government is failing Americans on AI | Shakeel Hashim

Source: The Guardian | Published: Sep 16 | Risk: MEDIUM | Impacted: Companies developing or integrating AI, Security and IT governance teams, CISOs overseeing AI deployments | Topics: Threat Actor / Ai

What happened: In his article, Shakeel Hashim criticizes the U.S. government’s lack of action on AI regulation, highlighting industry leaders’ calls for a development slowdown due to escalating risks. He argues that reliance on self-regulation by companies is insufficient and that government intervention is necessary to ensure public safety.

Why it matters: Without effective government oversight, organizations that rely on AI may face inconsistent standards, unchecked model abuse, and greater legal and reputational liabilities stemming from commercial or adversarial misuse of AI technology.

How it works: AI refers to software models capable of tasks requiring human-like decision making. Current regulatory gaps leave organizations to self-govern AI use, increasing risk of deviant behaviors or compliance failures as model adoption accelerates.

Practitioner Perspective

The current lack of regulatory direction on AI places the burden of risk assessment and governance on security and compliance teams. This vacuum makes it more difficult to set policies for model use, third-party AI integration, and incident response to model failure or misuse. Security leads must anticipate that regulatory requirements could be introduced with little notice, so defensible logging and explainable AI controls should be prioritized. Weak oversight also heightens the risk of adversarial behaviors (prompt injection, model inversion), which defenders must plan for proactively. Stay plugged in to policymaker movements as they will shape operational risk and resource allocation.

Recommended Actions

• Review AI-related risk acceptance procedures and ensure they address abuse and privacy scenarios specifically related to commercial models
• Track federal and state AI regulation proposals impacting your organization’s AI usage

‘Pacing’ won’t eliminate the risk of AI doom. Here’s what could | David Krueger

Source: The Guardian | Published: Sep 16 | Risk: MEDIUM | Impacted: AI development teams, Organizations depending on advanced AI infrastructure, CISOs managing AI risk portfolios | Topics: Ai

What happened: David Krueger argues that Dario Amodei’s ‘pacing’ proposal, which suggests a temporary slowdown in AI development, is insufficient to mitigate existential risks. He advocates for an immediate, indefinite international moratorium on frontier AI development, including decommissioning advanced computer chips and dismantling the supply chain for producing such chips, to effectively prevent potential AI-induced catastrophes.

Why it matters: Industry divides on the pace of AI development may produce unpredictable regulatory and market responses, affecting how risk managers must prioritize AI-related threat scenarios and crisis planning.

How it works: Frontier AI development involves using large-scale computer chips and supply chains to produce highly capable models. Policy debates are emerging around controlling or decommissioning these resources to manage existential risks.

Practitioner Perspective

This ongoing debate signals that defenders cannot rely on a stable external policy environment to dictate how aggressively they invest in AI safety controls. Calls for outright moratoriums or radical supply chain controls could shake technology roadmaps overnight, especially for firms reliant on advanced AI compute. It is essential to maintain situational awareness on both regulatory and activist campaigns, since either could impact supply access or internal governance. Robust internal alignment on acceptable AI risks becomes critical as the regulatory pendulum swings. If your business model depends on scaling advanced AI workloads, have contingency plans for abrupt policy shocks.

Recommended Actions

• Monitor proposals for international AI development moratoriums and their likely impact on sourcing GPUs or cloud AI compute
• Scenario plan for continuity or risk transfer if high-end AI chip supply becomes constrained

Decoding America: Why won’t Donald Trump regulate AI? – podcast

Source: The Guardian | Published: Sep 16 | Risk: MEDIUM | Impacted: US-based SaaS developers integrating AI, Compliance teams at enterprises, Organizations operating cross-border AI services | Topics: Ai

What happened: In this episode of ‘Decoding America,’ hosts Jonathan Yerushalmy and Reged Ahmad discuss the lack of bipartisan support for regulating artificial intelligence, despite global concerns over its rapid development. They also cover the Supreme Court’s recent ruling on mail-in ballot restrictions and President Trump’s involvement in the Irish reunification debate.

Why it matters: Uncertainty or partisanship in national AI policy creates compliance ambiguity, making it harder for organizations to predict rules for deployment, monitoring, and liability of AI-powered products.

How it works: Regulation of artificial intelligence is subject to shifting national priorities and political polarization, affecting how compliance and security teams operationalize controls for AI-enabled workflows.

Practitioner Perspective

In every regulated sector, shifting government opinion on AI amplifies legal and business risk for CISOs and product owners. Rapid AI innovation without bipartisan guardrails can lead to awkward catch-up scenarios where rules are imposed post-breach or after harms occur. Security leads should ensure their teams document AI-specific risk decisions and maintain flexibility to adapt operational controls ahead of late policy conversion. Watch for sector-specific requirements possibly being introduced by state or international regulators, bypassing gridlock at the federal level. Anticipate and plan for regulatory obligations in highly visible or sensitive AI applications.

Recommended Actions

• Map all client-facing uses of AI and document associated regulatory and policy risks
• Track ongoing legislative debates at both federal and state levels on AI regulation

Wednesday briefing: Why tech companies might be only too happy for us to believe AI will ‘kill us all’

Source: The Guardian | Published: Sep 16 | Risk: MEDIUM | Impacted: Buyers of commercial AI solutions, Security budget holders at large enterprises, Policy and standards teams | Topics: Ai

What happened: In today’s newsletter: It is hard to tell fact from fiction when it comes to AI. What is really going on – and what should the government do about it? Good morning. As a general rule, it pays to be suspicious of any gigantic company that claims it’s developing a tool capable of destroying humanity. But in recent days, a

Why it matters: Manipulation of AI risk narratives by major vendors may influence regulatory outcomes, investment, and the prioritization of security spending in ways that do not align with true threat models.

How it works: Tech companies may use communication strategies around AI risk to influence policy and industry standards, which can impact customer budgeting and regulatory pressure on buyers of AI solutions.

Practitioner Perspective

Security leaders cannot blindly accept vendor narratives about AI’s existential risk, especially if such stories may serve market or lobbying objectives. Defensive investment should remain grounded in concrete organizational risks, data privacy breaches, model poisoning, and adversarial ML techniques, rather than speculative threats. Maintaining independence in risk assessment ensures resources are allocated to actual exposures. Stay alert to subtle shifts in standard-setting that might advantage entrenched vendors at the expense of open security research or customer choice. Your job is to balance vigilance against overreaction dictated by outside interests.

Recommended Actions

• Scrutinize AI vendor marketing and public positioning for exaggeration or self-interested messaging
• Anchor internal AI security decisions in threat modeling and documented exploit scenarios

The Spin | Robots bowling the perfect doosra are some way off but AI is already reshaping cricket

Source: The Guardian | Published: Sep 16 | Risk: LOW | Impacted: Sports analytics vendors, Professional sports teams using AI, Partners integrating EquiPlay or similar platforms | Topics: Ai

What happened: Artificial intelligence is increasingly integrated into cricket, enhancing player analysis and performance. Analysts like Freddie Wilde utilize AI to process data more efficiently, aiding in strategic decisions. Additionally, platforms such as EquiPlay employ AI to standardize assessments and track player development, marking a significant shift in coaching and talent evaluation.

Why it matters: Adoption of AI-driven analytics platforms in professional sports introduces potential risks of data integrity loss or manipulation, especially where proprietary or sensitive performance data may have commercial or reputational value.

How it works: AI in sports is used for analyzing large volumes of player and game data, helping standardize evaluation and improve training. Performance depends on the accuracy and security of both the underlying data and the model algorithms.

Practitioner Perspective

Organizations in sports technology should treat AI model pipelines and associated data collection systems as attack surfaces. AI-fueled analytics can create competitive advantage, but if the input or models are tampered with, teams may lose their edge or face scandals. Technical defenses should include monitoring access to raw data, controlling integration points, and validating model training integrity. Datasets used for player assessment are assets that may attract criminal or insider interest. Ensure contracts with analytics providers require rigorous controls over data residency, model audit, and error reporting.

Recommended Actions

• Assess vendor controls for platforms like EquiPlay to ensure data integrity and minimal exposure
• Review access management to AI-driven sports analytics systems and datasets

I used to think AI would kill us all, until the techbro CEOs said AI will kill us all and now I’m not so sure | First Dog on the Moon

Source: The Guardian | Published: Sep 16 | Risk: LOW | Impacted: Corporate comms and security awareness teams, Boards facing AI investment decisions, Organizations exposed to media narratives on AI risk | Topics: Ics Ot / Ai

What happened: The cartoon humorously reflects on the paradox of tech CEOs warning about AI’s potential to harm humanity, leading to uncertainty about AI’s true threat.

Why it matters: Conflicting public narratives about AI-generated risk can exacerbate uncertainty for board members, regulators, and end users, making it harder for security teams to set expectations or justify investment in meaningful controls.

How it works: Public discussion of artificial intelligence risk often oscillates between catastrophic and skeptical positions, affecting stakeholder perception of technology deployments inside organizations.

Practitioner Perspective

Defenders must be aware that the AI ‘doomsday’ discourse is increasingly driven both by hype and genuine AI safety concern. Strategic communications planning is essential to prevent knee-jerk reactions among leadership or stakeholders. Security teams can’t afford to ignore reputational risk from either overstatement or underestimation of AI-related threats. Balance sober technology risk assessment with transparency about knowns versus unknowns in organizational AI use. The most critical task is to maintain clarity and trust during periods of public anxiety or contradictory expert messaging.

Recommended Actions

• Brief board and C-suite with fact-based AI risk overviews, explicitly separating technical risk from media narratives
• Prepare communication templates that address both existential and practical AI risk scenarios for stakeholders

Also Today

• Anthropic lands deal in $31bn datacentre in western Queensland, David Crisafulli says: Premier describes deal as a major win, boosting both jobs and state energy.
• Your photos, your words and your work: why is Labor making it easier for AI companies to take them for free? | Holly Rankin: Leaked proposal may shift copyright to an opt-out model, exposing online content to AI training without explicit consent.
• Anthropic CEO renews call for AI slowdown as Nvidia’s urges acceleration: At a major conference, Anthropic and Nvidia CEOs disagree on whether AI development should pause or push ahead.
• Roundtables: Could AI really kill us all?: Industry and academic leaders debate if advanced AI poses existential risks or if fears are overblown.
• The Download: AI doomers, whistleblowing agents, and de-aged livers: Leading AI executives now align on extinction risks, driving talk of urgent policy changes.

Defensive Actions

• Immediately patch all Marimo systems for CVE-2026-39987 or apply vendor-recommended mitigations
• Hunt for evidence of CVE-2026-39987 exploitation and subsequent SSH access attempts in SIEM/EDR logs
• Audit connectivity from Marimo nodes to SSH bastions and restrict with explicit allowlists
• Enforce MFA and strict SSH configuration (disable password authentication) for all bastion hosts connected to Marimo systems
• Rotate SSH keys on any systems potentially exposed via Marimo RCE
• Review AI-related risk acceptance procedures and ensure they address abuse and privacy scenarios related to commercial models
• Track federal and state AI regulation proposals impacting your organization’s AI usage
• Embed logging and explainability in deployed AI systems to prepare for regulatory changes

What We’re Watching

• Ongoing exploitation attempts of Marimo CVE-2026-39987 in the wild, patch status and threat actor targeting
• US and state-level developments in AI regulation with potential for sudden compliance mandates
• Upcoming policy statements or restrictions regarding advanced AI chip supply chains following recent calls for moratoriums
• Public and corporate response to increased senior AI leadership consensus around existential risks
• Evidence of data scraping or copyright conflicts as governments consider opt-out models for AI model training

Found this briefing useful? Follow the blog to get the next one as soon as it is published, and pass it along to a colleague who owns patching.

Categories: Artificial Intelligence, Cybersecurity Blog

Tags: AI policy, AI Security, exploits, RCE, Regulation