News

Today, we are excited to announce a major new capability for MapleNode: Tamper-Proof Memory powered by the Cognitum One Seed.

In a world increasingly driven by autonomous systems, AI agents, and machine-generated decisions, trust is becoming infrastructure. In this update we demonstrate how we turned the https://cognitum.one/">Cognitum Seed into a witness based trust edge device.

https://canxp.ai/maplenode">MapleNode is a sovereign edge cognition appliance for MapleOS. MapleNode features:

1. Local-first memory. Vector-searchable semantic store, on-device, with no cloud round-trip on the hot path.

2. Pluggable embeddings. sentence-transformers/all-MiniLM-L6-v2 (384-d) by default, with a deterministic hash fallback so the API stays up even if heavy ML deps fail to install.

3. Discoverable. mDNS publishes maplenode.local, and the node broadcasts its capabilities on UDP :8766 every 30s so MapleOS can find it without manual config.

MapleNode is not a GPU server. It is an edge memory + orchestration node. After developing MapleNode we realized one critical component of an appliance like this was the ability to "proof" memory.

Integration

Our MapleNode Edge Appliance is connected to the Cognitum Seed via a USB-Ethernet connection only. We disabled WIFI capability on the Seed to ensure it would not have any secondary network connections. Connect the micro-usb to the Seed and the USB3 to the Armbian Pi running MapleNode.

You will need to connect your Edge Appliance device running MapleNode with a functional Cognitum Seed.

This new integration gives MapleNode the ability to create cryptographically verifiable memory records, allowing organizations to prove that AI memory, events, workflows, and knowledge entries have not been modified after creation.

How does it work?

When a Cognitum Seed is paired and MAPLENODE_WITNESS_MODE != "off", every write through /memory/store, /documents/upload, and /documents/index generates a canonical digest of the record's identifying fields, asks the seed's hardware Ed25519 key to sign that digest, and persists the signature in SQLite alongside the row. Any future caller can ask maplenode to re-verify by recomputing the same digest from the stored row and asking the seed whether the stored signature is valid for it.

A bare vector database (SQLite + LanceDB) gives us semantic recall but not integrity. Anything with write access to maplenode.db can mutate stored text after the fact, change a memory's content, alter a metadata field, re-point a chunk at a different document, and nothing in the database betrays that change.

Witness-stamping closes that gap for memories written while a seed was paired: every such memory carries an Ed25519 signature over its canonical form. After-the-fact tampering breaks the signature.

It is not full-blown end-to-end encryption, and it is not an immutable ledger. It is a cryptographic receipt that the row's identifying content was attested by hardware at write time.

For more information please read our https://github.com/CanXPAI/maplenode/blob/main/docs/witness.md">Witness documentation on this effort.

Why Tamper-Proof Memory Matters

Most AI systems today operate like black boxes. An AI agent performs a task. A workflow executes. A decision is made. A memory is stored. But over time, organizations lose confidence in what actually happened, when it happened, and whether records were altered afterward.

That is a massive problem for enterprise AI. Legal firms need verifiable evidence chains. Healthcare organizations need trusted audit trails. Industrial operators need immutable operational histories. Defence and public-sector deployments require provable integrity. Even standard enterprise automation increasingly depends on trustworthy event histories.

Traditional databases were never designed for AI memory integrity. Tamper-proof memory changes that. By integrating the Cognitum One Seed into MapleNode, memory entries and operational events can now be cryptographically witnessed and anchored into a verifiable chain of trust. This creates a provable record that memory existed in a specific state at a specific moment in time.

This is "Proof of Trust" in a tamper-proof form.

MapleNode as a Witnessed AI Edge Appliance

MapleNode was already designed as a sovereign edge AI appliance capable of running private models, local inference, secure knowledge systems, and AI orchestration closer to the organization. Now it gains something even more important: verifiable memory integrity.

This transforms MapleNode from a private AI appliance into a witnessed intelligence system.

AI workflows running on MapleNode can now generate immutable evidence trails for:

• AI agent actions

• Memory writes

• Knowledge ingestion

• Operational events

• Workflow execution

• Human approvals

• Audit checkpoints

• Chain-of-custody records

• Compliance and governance events

For regulated industries, this is a major solve. Organizations no longer need to blindly trust that AI systems behaved correctly. They can independently verify what occurred.

Sovereign AI Requires Verifiable Infrastructure

At CanXP AI, we have repeatedly said that sovereignty is not just about where models are hosted.

It is about:

• Who controls the infrastructure

• Who governs the data

• Who owns the models

• Who can verify the system

• Who can trust the memory

Tamper-proof memory is part of building sovereign AI systems that organizations can actually rely on. As AI moves deeper into healthcare, law, industrial operations, government workflows, and critical infrastructure, auditability becomes essential.

The Edge Is Becoming Intelligent and Accountable

One of the most important shifts happening in AI is the movement away from centralized black-box cloud systems toward distributed, regional, and edge-based intelligence. MapleNode represents this transition. Instead of sending every workflow to a foreign cloud platform, organizations can deploy AI systems locally, closer to their teams, facilities, clinics, and operational environments.

Now, with witnessed tamper-proof memory, those edge systems gain accountability.

This is especially important for:

• Healthcare AI systems

• Legal AI assistants

• Industrial automation

• Defence and dual-use AI

• Scientific and research environments

• Regional government deployments

• Human-in-the-loop operational AI

In these environments, memory integrity matters just as much as model performance.

Building the Foundations of Trusted AI

The AI industry has spent years racing toward larger models and bigger compute clusters. But the next generation of AI infrastructure will be defined by something else entirely: Trust.

Can organizations trust what an AI system remembers?
Can they verify what occurred?
Can they audit workflows after the fact?
Can they prove data integrity?
Can they establish cryptographic chain-of-custody for machine operations?

MapleNode’s new Cognitum One Seed integration is part of answering those questions. This effort aligns with CanXP AI’s broader mission to build sovereign, privacy-first, human-centered AI infrastructure that organizations can govern, verify, and deploy on their own terms. Because in the era of autonomous systems, memory itself becomes critical infrastructure. Critical infrastructure must be trustworthy.

For more information about MapleNode, visit: https://canxp.ai/maplenode">https://canxp.ai/maplenode

Additional technical details about the tamper-proof witness architecture can be found in the MapleNode witness documentation on https://github.com/CanXPAI/maplenode/blob/main/docs/witness.md">GitHub.

Stay tuned. In the coming updates we will be explaining how users can use MapleNode (with or without the Seed) with MapleOS.

The small Cessna Caravan accelerates down the runway and climbs into the air, all while the pilot beside me keeps his hands off the controls.

“Let’s see those jazz hands,” jokes Tim Burns, chief technology officer at startup Merlin Labs, over the airplane’s intercom from a back seat.

On this flight, test pilot Matt Diamond in the left seat beside me is not controlling the airplane at all. Many of the normal tasks of piloting are instead being handled by artificial intelligence.

I am, legally speaking, a test subject — even the airplane is labeled “experimental.” The Merlin Pilot system handles much more than a traditional autopilot, using a natural language processing model to listen to instructions from a mock air traffic controller and responding over the radio using a computerized female voice. Test pilot Diamond says, “Authorize,” and the airplane begins turning to a new course.

As a pilot myself — and admittedly a bit of a control freak — surrendering control to a computer did not come naturally. But the demonstration is an important one as more aviation companies are looking to AI to usher in a new evolution in air travel by using it to automate tasks for pilots and perhaps one day enable fully autonomous.

Our flight is taking place as airlines worldwide are facing a growing pilot shortage. Boeing estimates that carriers will need more than 600,000 new pilots over the next two decades. At the same time, aviation safety officials are confronting increasing pressure on an already strained air traffic control system following a series of high-profile close calls and deadly accidents in recent years.

The push toward AI-assisted aviation is also gaining support in Washington. Transportation Secretary Sean Duffy has promoted artificial intelligence tools as part of the Trump administration’s broader push to modernize the nation’s aging air traffic control system.

“We are never going to outsource the national airspace to AI tools,” Duffy told CNN in a recent interview. “Controllers are going to control the airspace, but we can make their jobs easier.”

Duffy said the administration sees AI as a way to reduce workload for controllers and improve efficiency across increasingly crowded airspace.

Merlin argues artificial intelligence could eventually help address some of the same problems in the cockpit. “Eighty percent of accidents in aviation are still caused by human error,” Merlin CEO Matthew George told CNN. “If we can reduce that, that’s a pretty useful way to spend our time.”

The idea remains controversial. Commercial aviation has steadily added automation for decades, leading to today’s fly-by-wire systems in which computers interpret pilot inputs even during manual flight.

“Modern cockpits have quite a bit of automation already, but the automation is within a narrowly defined scope,” said Mykel Kochenderfer, whose research at Stanford University focuses on autonomous systems and aviation safety. Kochenderfer said newer AI-assisted systems are designed to handle a broader range of unexpected situations than traditional rule-based automation.

“Our experience shows this can be a very promising way to enhance safety,” he said, “but the industry has a long way to go to further harden the technology and establish the trust required for acceptance.”

Changing the minds of pilots might not be easy. Current in-flight automation systems place the pilot at the center, allowing them to intervene when necessary.

Capt. Jason Ambrosi, president of the Air Line Pilots Association which represents more than 79,000 pilots in the United States and Canada, says automation and AI should support pilots, not replace them.

“Technological advancements can improve aviation safety, but they will never be a substitute for the pilots on an aircraft,” Ambrosi said in a statement to CNN. “The most important safety feature on every airline flight will always be two well-trained and rested pilots on the flightdeck.”

Merlin underscores fully pilotless passenger flights are still far away. “We’re not flipping a switch to uncrewed airplanes,” George said. “This is about putting AI alongside human pilots and building trust.”

The company says it has completed hundreds of test flights as it works toward certification from the Federal Aviation Administration. Those standards are among the strictest in transportation, often requiring years of testing and redundancy analysis before new systems are approved.

The military may provide the system’s first major proving ground. Merlin recently secured a contract worth more than $100 million with the US Air Force to eventually bring the technology to C-130 cargo planes.

As the Merlin system lines us up on final approach, it starts a gradual descent toward runway 34 and jockeys the controls to stay on the flight path, despite a slight crosswind, all the way to touchdown.

“It’s a challenging problem for the automation,” test pilot Diamond says to me as we are taxiing back to Merlin’s hangar. “But once you crack it, it makes things much easier on the pilot.”

Source: https://www.cnn.com/2026/05/24/us/ai-flying-airplanes">https://www.cnn.com/2026/05/24/us/ai-flying-airplanes

For the last two years, much of the AI industry has been selling a comforting illusion: powerful intelligence, available instantly, for the price of a streaming subscription. When we began building CanXP AI this issue was always top of mind. Which is why we decided to move away from that misleading subscription model because it's not what businesses and organizations actually need. The $20/month subscription model was a lie being told by our biggest competitors like OpenAI and Anthropic. Now their largest customers are starting to realize this.

Twenty dollars a month. A few clicks. A chatbot in every workflow. A magic assistant for every employee. It was great while it latest. Now what? Now we need the next era of AI. Now we need what CanXP AI has spent nearly 2 years developing in stealth. Now we need to give everyone the ability to build their own model based on their data and inference - not the frontier. The frontier is great for the big generalizations but for the day to day grind - we don't recommend it unless you have a very big bank account. Bigger than Microsoft's even.

Microsoft is reportedly canceling most internal Claude Code licenses and moving many engineers back toward GitHub Copilot CLI, with The Verge reporting that the decision is partly financial and partly about giving Microsoft more control over its own engineering workflows, security expectations, and repositories. Uber, meanwhile, has already felt the other side of the AI productivity boom: Business Insider reported that Uber’s CTO said the company had already spent its 2026 Claude Code budget, while the CFO acknowledged the company had underestimated the impact and adoption of AI tools.

These are not stories about AI failing. They are stories about AI succeeding so aggressively that the old pricing model collapses. Organizations that were counting on it are in for a rude awakening.

Once AI moves from occasional chat usage into daily, agentic, company-wide execution, the economics change completely. The cost is no longer a simple monthly seat. The cost becomes usage, tokens, context windows, tool calls, codebase scans, file reads, retries, orchestration loops, and every automated step taken on behalf of the user.

The future of enterprise AI will not be defined by who can rent the largest frontier model forever. It will be defined by who can own, train, host, govern, and deploy the right model in the right place.

That is where self-trained small language models matter. This is where CanXP AI enters the room.

The Enterprise Lesson we Learned: Frontier Models Are Powerful, But Dependency Is Expensive

Claude, GPT, Gemini, Cohere, and other frontier-class models are extraordinary technologies. They are excellent generalists. They can reason across domains, write, code, summarize, plan, and assist with broad creative or technical tasks.

But broad capability is not the same thing as enterprise efficiency.

Most organizations do not need a trillion-parameter-scale model thinking from first principles every time someone asks a repeatable business question. A hospital does not need a frontier model to understand the same clinical workflow 40,000 times. A law firm does not need to send confidential contract patterns through a remote API forever. A manufacturer does not need every maintenance log analyzed by the most expensive model available. A defence contractor does not need foreign-hosted inference touching sensitive operational material.

The chart https://x.com/HedgieMarkets/status/2057531661785628841">posted on X by HedgieMarkets tells the story clearly: the future is not “one giant cloud model for everything.” The future is a layered AI architecture.

Frontier models still matter. But they should sit at the top of the stack for high-complexity reasoning, research, synthesis, and exception handling. Underneath that, organizations need self-trained small and medium language models that understand their terminology, documents, methods, workflows, policies, customers, and operational constraints.

That is how AI becomes infrastructure instead of a monthly addiction.

Self-Trained SLMs Change the Economics

The strongest argument for self-trained SLMs is not ideological. It is practical.

A self-trained small language model can be trained or fine-tuned on a specific corpus, task, domain, methodology, or workflow. Once trained, it can be hosted on private cloud infrastructure, on-prem servers, edge devices, workstations, or even optimized consumer hardware depending on the workload.

That changes the cost curve. Instead of paying endlessly for every token sent to a third-party frontier model, the organization can amortize training and hosting costs across repeated use. For high-volume workflows, this matters enormously. The more the model is used, the more valuable ownership becomes.

This is why the “$20/month AI” story was always misleading. Twenty dollars a month works for casual individual usage. It does not survive when AI becomes an operational layer across engineering, healthcare, legal review, industrial analysis, internal support, compliance, and data processing.

Self-Trained Models Understand the Work, Not Just the Prompt

RAG, prompt engineering, and API guardrails are useful. But they are not the same thing as training. CanXP AI provides full AI Surface support via our MapleOS. You can build Surfaces which are applications with their own specific harness engineering that can save alot of money.

When an organization relies entirely on a frontier model with a retrieval layer, it is often trying to patch domain knowledge onto a model that was never trained to deeply understand the organization’s specialized methodology in the first place.

That works for simple retrieval. It fails when the task depends on judgment, sequence, procedure, language, constraints, or expert-specific reasoning.

A self-trained SLM can internalize the patterns of a domain more directly. It can learn how a medical practice structures assessments. It can learn how a law firm reasons through contract risk. It can learn how an industrial operator categorizes failures. It can learn the vocabulary of a defence program, the structure of a scientific workflow, or the recurring judgment patterns of a professional service organization.

That does not mean the model becomes infallible. It means the intelligence is moved closer to the work.

That is the real breakthrough.

On-Device and On-Prem AI Are Not Nostalgia. They Are Strategy.

For years, the industry pushed everything to the cloud. That made sense when models were too large, hardware was scarce, and inference required specialized infrastructure.

But the next phase of AI is different.

Quantization, smaller open-weight models, better inference engines, NPUs, local GPUs, edge devices, and private deployment tooling are making local AI practical again. Not for every use case. Not for every model. But for a growing number of enterprise workflows, on-device and on-prem inference is not only possible, it will be preferred.

CanXP AI’s own sovereign AI infrastructure work already supports on-device inference as a strategic differentiator for privacy, cost, reliability, sustainability, and reduced cloud API dependency.

This matters especially in Canada.

Healthcare, legal, government, defence, industrial, and scientific AI cannot be treated like consumer chatbot usage. Jurisdiction matters. Provincial rules matter. Sensitive data matters. Governance matters. If the model is making recommendations, summarizing confidential material, assisting clinical workflows, or processing regulated information, then the location of inference is not a technical footnote. It is a core requirement.

The Future Is Hybrid AI

Hybrid is the architecture that serious organizations will be forced to adopt as AI usage scales.

Microsoft’s move is revealing because Microsoft is not walking away from AI. It is doing the opposite. It is consolidating control around tooling it can shape, govern, integrate, and optimize for its own workflows. That is exactly the lesson every enterprise should take from this moment.

If Microsoft wants control over its AI tooling, why would a hospital, law firm, manufacturer, government department, or Canadian enterprise want permanent dependency on foreign-hosted black-box APIs? We don't think so.

The Winners Will Own Their Models

The first wave of AI adoption was about access.

The next wave is about control.

The winners will not be the companies that simply give every employee another AI subscription. The winners will be the organizations that identify which workflows deserve frontier intelligence, which workflows deserve a trained internal model, which workloads must run locally, and which data should never leave their environment.

Self-trained SLMs are not a downgrade from frontier AI. They are the specialization layer that makes AI economically and operationally sustainable. They are how professionals move from prompting around a model’s ignorance to training models that understand the work. They are how regulated industries adopt AI without surrendering control. They are how Canadian organizations build AI capacity under Canadian governance. And they are how the enterprise escapes the trap of renting intelligence forever. The good ole days of cheap, unlimited, $20/month AI is a dead end.

Feel free to join us and begin building the next era of AI on your terms. Not a frontiers.

Microsoft drops Claude Code for thousands of internal engineers and is moving them to GitHub Copilot CLI by June 30, 2026 — even though Anthropic's tool was the more popular pick. You'll see why the timing matches the end of Microsoft's fiscal year and what this tells us about enterprise AI in 2026.

• Microsoft is canceling thousands of internal Claude Code licenses across its Experiences + Devices group — the team behind Windows, Microsoft 365, Outlook, Teams, and Surface — with a June 30, 2026 cutoff.

• Engineers are being moved to GitHub Copilot CLI, Microsoft's own command-line AI coding tool, even though many of them prefer Anthropic's Claude Code.

• The official reason is "toolchain unification," but the timing lines up with the end of Microsoft's fiscal year, which means cost-cutting is part of the story too.

• Anthropic's models aren't being banned at Microsoft. Claude will still be available inside Copilot CLI. Only the Claude Code interface is going away internally.

Microsoft just told thousands of its own engineers to stop using one of the most popular AI coding tools on the market — Anthropic’s Claude Code — and switch to its own GitHub Copilot CLI by June 30. The shift, first reported by The Verge’s Tom Warren, comes six months after Microsoft handed those same engineers Claude Code seats and watched adoption explode. Here’s what changed, why it matters for the AI tools you might use yourself, and what it says about the bigger fight over who controls the AI coding stack.

Source: https://memeburn.com/microsoft-drops-claude-code-what-changes-for-you/">https://memeburn.com/microsoft-drops-claude-code-what-changes-for-you/

If you are wondering what the impact of relying too heavily on frontier models like Claude. Look no further than Uber. Uber exhausted its entire 2026 artificial intelligence budget by April, four months into the calendar year, after Anthropic's Claude Code spread across roughly 5,000 engineers faster than the company's finance models had anticipated. Chief Technology Officer Praveen Neppalli Naga https://www.theinformation.com/newsletters/applied-ai/uber-cto-shows-claude-code-can-blow-ai-budgets">confirmed the overrun to The Information, saying the company was back to the drawing board on its assumptions. Uber's total research and development spend reached https://www.benzinga.com/markets/tech/26/04/51828848/ubers-anthropic-ai-push-hits-wall-cto-says-budget-struggles-despite-spend">$3.4 billion in 2025, up 9 percent year over year, which makes the budget collapse less about scale and more about a pricing model that enterprise finance teams have not learned how to manage. This is why we created our new SLM training facility. We don't know exactly what Uber was doing with those tokens but we can only assume it was for coding, management, and operations. Using Claude for some of those tasks is like using a sledge hammer to mount a wall tack.

The disclosure landed alongside a structural shift from Anthropic itself. On May 13, the company https://www.axios.com/2026/05/14/anthropic-claude-price-openai-tokens">announced that paid Claude subscribers would soon face a separate monthly credit meter for agent tools and third-party harnesses, billed at full application programming interface rates starting June 15. Read together, the two events describe a single problem. Token-based consumption pricing does not behave like the software line items chief financial officers know how to model, and the gap between what engineers consume and what finance teams expect is no longer hypothetical.

How A Coding Tool Outran A Budget

Uber rolled out Claude Code to its engineering organization in https://startupfortune.com/uber-burned-its-entire-2026-ai-budget-in-four-months-and-claude-code-is-why-finance-teams-should-be-worried/">December 2025. Adoption climbed from 32 percent of engineers in February to 84 percent classified as agentic coding users by March. By spring, https://byteiota.com/uber-blows-2026-ai-budget-on-claude-code-in-4-months/">95 percent of Uber engineers used artificial intelligence tools monthly, and roughly 70 percent of committed code originated from those tools. About 11 percent of live backend updates were written by agents with no human in the loop, according to Uber's own disclosures.

The numbers behind the spend are what make the story instructive rather than anecdotal. Monthly cost per engineer ranged from $150 to $250 on average, with power users running between $500 and $2,000. Naga himself reported https://startupfortune.com/uber-burned-its-entire-2026-ai-budget-in-four-months-and-claude-code-is-why-finance-teams-should-be-worried/">spending $1,200 in a two-hour session during a personal demo. The tool did not fail, and engineers did not misuse it. They used it for exactly the workloads it was designed to handle, parallel agent execution, large-scale codebase refactoring, automated test generation and backend code production. From a productivity standpoint the rollout was a success. From a finance standpoint it was a runaway.

Uber compounded the dynamic by ranking engineers on internal leaderboards based on Claude Code usage. That created a cultural incentive to consume more tokens, which translated directly into faster budget burn. The teams driving adoption were not the same teams managing the spend, and that organizational gap turned out to be the load-bearing flaw.

Source: https://www.forbes.com/sites/janakirammsv/2026/05/17/uber-burns-its-2026-ai-budget-in-four-months-on-claude-code/">https://www.forbes.com/sites/janakirammsv/2026/05/17/uber-burns-its-2026-ai-budget-in-four-months-on-claude-code/

https://www.independent.co.uk/topic/linkedin">LinkedIn is suppressing posts and comments on https://finance.yahoo.com/news/linkedin-set-layoff-5-percent-175010171.html">the networking platform written with the help of https://www.independent.co.uk/topic/artificial-intelligence">artificial intelligence, the https://www.independent.co.uk/topic/microsoft">Microsoft-owned company’s vice president of product, Laura Lorenzetti, announced on Wednesday.

Posts flagged as https://tech.yahoo.com/ai/articles/more-third-podcasts-now-ai-201703125.html">AI-generated by its new detection system will not be removed, but will be suppressed from recommendations, LinkedIn said in a blog post.

The system could https://www.independent.co.uk/tech/ai-artifical-intelligence-tool-humans-writing-photos-b2865538.html">correctly flag generic https://www.independent.co.uk/topic/ai">AI-generated content 94 per cent of the time in early tests, the company said.

“We’re seeing a rise in what many call ‘AI slop’, content that is low-effort, AI-generated content that may sound polished on the surface but lacks any real unique perspective or substance,” Ms Lorenzetti said.

“At a time when more people need help navigating work, it’s more important than ever that people can learn from real voices, authentic perspectives, and lived expertise,” she wrote, adding that the ultimate value of posts “comes from the human behind the tool”.

The company said its new systems have been trained to recognise signals of AI slop, including “content that feels generic or repetitive, even if it appears polished on the surface”.

“When content appears to be generated by AI and lacks a clear perspective, it is less likely to be widely distributed beyond a person’s immediate network,” it said.

LinkedIn hopes users will see less of such generic AI content from outside one’s network on their feed.

In tests, users could already see fewer of these types of posts in their feeds, with the company expecting more of this feature to be experienced across the platform.

“Bots and fake AI profiles ruin genuine engagement,” LinkedIn said, adding that the new updates are “designed to protect the experience on LinkedIn so when you’re engaging, you’re interacting with real people offering their real point of view”.

However, the platform seems to be walking on a thin line in its application of AI, as it enables users to use its own feature called “Rewrite with AI” in its post composer.

Source: https://ca.news.yahoo.com/linkedin-cracks-down-ai-slop-045111486.html?guccounter=1&guce_referrer=aHR0cHM6Ly93d3cuZ29vZ2xlLmNvbS8&guce_referrer_sig=AQAAAHxe2jZMeCBGCySiABWf8kXdCtGHxpwqAg4AvUvYW8NO_2sIWDiFTRmkrAafl3Azk0QFY4NF_t0b_9JDxwA-Z8XItke_Mdidkq6IlcJi3wDX41-T7TSvPTBmoWY3FyHeONIP8h1_d899vDRFCRFx9tffc_IBpFcNG0dC-FD6DGin">https://ca.news.yahoo.com/linkedin-cracks-down-ai-slop-045111486.html?guccounter=1&guce_referrer=aHR0cHM6Ly93d3cuZ29vZ2xlLmNvbS8&guce_referrer_sig=AQAAAHxe2jZMeCBGCySiABWf8kXdCtGHxpwqAg4AvUvYW8NO_2sIWDiFTRmkrAafl3Azk0QFY4NF_t0b_9JDxwA-Z8XItke_Mdidkq6IlcJi3wDX41-T7TSvPTBmoWY3FyHeONIP8h1_d899vDRFCRFx9tffc_IBpFcNG0dC-FD6DGin