by admin | Oct 2, 2026 | Future Jobs
If you have been wondering what an “AI engineer” job will even look like a couple of years from now, Anthropic just gave a surprisingly detailed answer.
On October 2, 2026, Anthropic launched Claude Frontier Academy, a $100 million program built to train 10,000 engineers to deploy AI inside real companies by the end of 2027. It is not a course you can sign up for on a whim, and it is not aimed at beginners. But it tells us something useful about where enterprise AI jobs are actually heading, and it is worth understanding even if you never set foot in one of its classrooms.
What Is Claude Frontier Academy, Exactly?
According to Anthropic’s own announcement, Claude Frontier Academy trains what it calls “Frontier Deployed Engineers,” using a structure the company compares to a medical residency. Instead of a short certificate course, participants learn directly from Anthropic’s own engineers, work through realistic deployment cases, and have to pass a graded assessment before they earn a credential.
The goal is not to teach people how to chat with an AI model well. It is to teach experienced software engineers how to actually deploy Claude inside a large organization, safely and usefully, which turns out to be a much harder and more specific skill than most job listings make it sound.
How the Program Actually Works
The residency runs in five stages:
- A multi-day, in-person program working directly alongside Anthropic engineers
- A simulated enterprise deployment, including picking a use case and taking it through a security review
- A graded practical exam based on that simulated work
- A 12-week residency leading a real Claude project inside the participant’s own company
- A final assessment that leads to a credential
Pass the first stage and you earn a Claude Resident Engineer badge. Finish the full 12-week residency and pass the final review, and you earn the Claude Frontier Deployed Engineer badge. The first cohorts are running in San Francisco, New York, and London, with the first badges expected in early 2027.
Who Can Actually Get In
This is the part most people miss: you cannot apply to Claude Frontier Academy directly. Participation works by nomination, and your employer has to ask Anthropic’s account team or Partner Account Manager whether your organization is eligible. Anthropic says it is looking for hands-on software engineers with strong fundamentals, real experience building with large language models, and a track record of helping others adopt AI on business-critical problems. Prior experience with AI agents specifically is not required.
The first cohorts already include engineers nominated by Accenture, Bain, Capgemini, Commonwealth Bank of Australia, Deloitte, McKinsey, Morgan Stanley, and Novo Nordisk, which gives a pretty clear signal about the kind of large, consulting-heavy or regulated organizations this is built for first.
Tip: you do not need a nomination to start building toward this kind of role. From my own experience working across websites, online tools, and cybersecurity projects, the engineers who get picked for programs like this are almost always the ones who were already quietly doing the work, not the ones waiting for a formal invitation. Deploying a small AI tool for a real task at your own job, documenting what went wrong and how you fixed it, is the closest free substitute available right now.
Why This Matters Even if You Will Never Be Nominated
Claude Frontier Academy is a single company’s program, but it reflects a much bigger problem. The World Economic Forum estimates that AI and information processing will affect 86% of businesses by 2030, and roughly 1.1 billion jobs could be reshaped by technology over the next decade. The Forum’s own conclusion is blunt: AI will create more jobs than it destroys, but only if companies deliberately invest in training people, rather than just installing new software on top of old job descriptions.
That is exactly the gap Anthropic is trying to fill with its own residency model, and it lines up closely with what we covered in our look at McKinsey’s recent report on AI and career transitions: the jobs that disappear and the jobs that get created rarely look the same, and the people who prepare early have a real advantage. Our broader explainer on how AI is changing future jobs covers the same shift from a wider angle.
A new, named, credentialed job title like “Frontier Deployed Engineer” is also a useful signal on its own. When a company this size spends $100 million building a formal training pipeline for a role, it usually means that role is about to show up in a lot more job postings, even outside the original partner companies.
If you want to start building toward that kind of role on your own terms, our guide on how to build an AI portfolio for free walks through creating small, finished projects that prove you can actually do the work, and our breakdown of what AI skills are actually worth on your CV shows how specific, demonstrable claims outperform vague ones when recruiters are deciding who to call.
Common Questions
Can I apply to Claude Frontier Academy myself?
Not directly. Participation is by nomination only. Your employer has to contact its Anthropic account team or Partner Account Manager to check eligibility.
What exactly is a “Frontier Deployed Engineer”?
It is Anthropic’s name for an engineer trained and credentialed to safely deploy Claude inside a real company, covering everything from picking a use case to passing a security review and running a live 12-week project.
Do I need prior experience with AI agents to qualify?
No. Anthropic says strong software engineering fundamentals and real large language model experience matter more than prior agent-specific work.
Will this program expand beyond the first eight partner companies?
Anthropic has not published a specific list of who comes next, but with a $100 million commitment and a target of 10,000 trained engineers by 2027, further partner organizations are a reasonable expectation.
Final Takeaway
Claude Frontier Academy will stay out of reach for most readers, at least for now. But it is a useful preview of where serious AI jobs are heading: toward named roles, formal credentials, and real accountability for getting deployments right, not just knowing how to write a good prompt. Whether or not a nomination ever lands in your inbox, the underlying skills it is built around, hands-on deployment experience and the ability to show your work, are ones you can start building today.
by admin | Oct 1, 2026 | Research & Productivity
If you have ever typed a research question into Google Scholar and gotten back a wall of a hundred papers with no idea where to start, you are not alone. That gap between “I have a question” and “I have the right ten papers to read” eats up more time than almost anything else in research work. Google has been quietly trying to close that gap with a tool called Google Scholar Labs, and it just got a big upgrade.
If you are a student, a grad researcher, or anyone who reads academic papers for a living, this is worth five minutes of your time.
What Is Google Scholar Labs?
Scholar Labs is an AI search mode built into Google Scholar. Instead of just matching keywords, it reads your actual question, works out the different angles hiding inside it, and then searches across Scholar to find papers that address those angles together.
Google gives a simple example: if you ask about caffeine’s effect on short-term memory, a keyword search gives you papers on caffeine and separate papers on memory. Scholar Labs tries to surface papers that connect the two, and for each result it explains how that paper actually answers your question.
It first launched in November 2025 to a small group of logged-in users. You can try it directly at scholar.google.com/scholar_labs/search.
What Changed in the Latest Update
In June 2026, Google rolled out a significant update to Scholar Labs, and the numbers are the headline:
- Results return about 10 times faster than the original version
- The tool scans up to 3 times more papers per search
- Logged-in users can now run 10 times more searches per day
Just as important, access widened from a small test group to all logged-in Google Scholar users. The update launched in English first, with other languages planned.
You can also ask follow-up questions once you get results, which turns a single search into something closer to a conversation with your reading list. That matters if your first question was too broad or you need to chase a specific detail, like a sample size or a methodology, across several papers.
How to Use It for Your Own Research
Getting value out of Scholar Labs comes down to how you phrase the question, not just what topic you search.
A plain keyword search still works fine when you already know the paper or author you want. Scholar Labs earns its keep when your question has more than one moving part. Try something closer to “how does remote work affect employee mental health across different industries” instead of just “remote work mental health.” The more structure you put into the question, the more structure you get back in the answer.
A few practical ways to use it:
- Early-stage literature review, to get a first map of a topic before you commit to a direction
- Checking whether your research question has already been well covered, before you propose it
- Verifying a claim or citation while you are drafting, by asking the specific question your sentence depends on
Important tip: treat Scholar Labs as a faster way to find candidate papers, not as a replacement for reading them. It is still an experimental AI feature, and the usual rules apply: check the actual paper, check the date, and check who funded or published it before you cite anything.
From working on websites and digital tools for a few years now, the tools that stick are rarely the flashiest ones. They are the ones that remove one annoying step from a task you already do every week. Scholar Labs fits that description for anyone doing a literature review: it does not replace your judgment, it just gets you to the reading list faster.
Where This Fits With Other AI Research Tools
Scholar Labs is one piece of a bigger shift toward AI-assisted research. If you want the wider picture, What Is AI Deep Research? explains how tools like this fit into a full research workflow, and OpenAI Built an AI Research Intern covers a similar idea from a different company.
If you are already using reference managers, Zotero’s new read-aloud AI feature is a good companion tool once Scholar Labs has helped you find the papers worth reading closely. And once you have your reading list, this guide to summarizing research papers with AI walks through getting through them faster without skipping the important parts.
For the official details straight from Google, see the Google Scholar blog’s launch post and the June 2026 update announcement.
Common Questions
Is Google Scholar Labs free to use?
Yes. It is available to any logged-in Google account holder through Google Scholar, with no separate subscription.
Does it replace Google Scholar’s normal search?
No. Scholar Labs is a separate AI mode you access through its own page. Regular Google Scholar search still works exactly as before.
Can I trust the papers it recommends?
Treat it the same way you would treat any search tool’s results: a useful starting point, not a final answer. Always check the original paper before citing it.
Is it available outside English?
The June 2026 update launched in English first. Google has said other languages are planned, but no firm dates have been announced.
Final Takeaway
Google Scholar Labs will not write your literature review for you, and it should not. What it does is cut down the time between asking a real research question and finding papers that actually address it, which is where most of us lose hours we do not have. If you do any kind of academic or professional research, it is worth bookmarking and testing on your next real question.
by admin | Sep 30, 2026 | Future Jobs
If you have ever wondered whether your job will still exist in ten years, you are not alone. A new report from the McKinsey Global Institute, released this week, puts a real number on that worry: as many as 11 million American workers may need to change occupations entirely by 2035 because of AI.
That headline sounds alarming on its own. But the full report, called “Workforce in Motion,” is actually more useful than scary once you read past the number. It tells you which jobs are shrinking, which are growing, which skills are suddenly worth a lot more, and who is most at risk of getting stuck. This post breaks down what McKinsey found and what it means for your own AI job transitions, whether you are years into a career or just starting one.
What McKinsey’s Report Actually Found
McKinsey estimates that automation could reduce demand for roughly 36 million US jobs over the next decade. At the same time, growth in the AI value chain and the wider economy could create demand for more than 40 million jobs. Net result: the US economy could end up with about 5 million more jobs than it has today, not fewer.
The catch is that the jobs disappearing and the jobs appearing are rarely the same jobs, or even in the same city. About 11 million workers, roughly 7 percent of the current workforce, may need to move into a completely different occupation. McKinsey says that works out to around 770,000 people switching occupational groups every year through 2035, which is more than three times the historical average. For comparison, that pace is close to what happened during the pandemic, when job switching briefly spiked in a similar way.
The steepest declines are concentrated in three areas: office and administrative support, retail and sales, and transportation and logistics. On the other side, healthcare support, construction, management, manufacturing, education, and IT roles are all projected to grow.
Why This Isn’t a “Robots Take All the Jobs” Story
McKinsey partner Anna Kortis summed it up well: this shift will happen faster and at a bigger scale than past transitions, but it is not mainly about jobs vanishing. It is about roles getting rebuilt. The report estimates that about 70 percent of workers will see their current role change in some way, even if they never switch job titles. Around a quarter of workers will see major task changes, more than 30 percent of their work hours shifting to new kinds of tasks, while most others will see smaller adjustments.
From my own experience working across websites, online tools, and cybersecurity over the past few years, this tracks with what I have actually seen happen. Very few of the tools I use today replaced a job outright. What they did was quietly change what the job involves, so the people who adapted fastest ended up doing more interesting work, not less work.
The Skills That Are Suddenly Worth More
One of the most useful parts of the report is a breakdown of which skills are becoming more valuable. McKinsey groups them into three buckets:
- Essential skills: problem-solving, leadership, communication, and attention to detail, which matter across almost every occupation.
- Enabling skills: decision-making, innovation, and critical thinking, which show up disproportionately in higher-paying roles.
- Empowering skills: AI fluency, adaptability, resilience, and curiosity, the traits that let someone keep learning as tools keep changing.
The report found that demand for AI fluency specifically has grown 11 times since 2022. Demand for adaptability is up fivefold. If you want one practical place to start, our guide to the AI skills that matter most for future jobs walks through how to build these without needing a technical background.
Tip: you don’t need to become a programmer to benefit from this shift. McKinsey’s own data shows the fastest-growing skill demand is for AI fluency and adaptability, not coding. Knowing how to use AI tools well in your own field counts.
Who Faces the Hardest Path
This is the part of the report that deserves more attention than it has gotten. The transition is not landing evenly. Lower-wage workers are 7.6 times more likely to need an occupational change than higher-wage workers. Workers without a bachelor’s degree are 1.8 times more likely, women are 1.6 times more likely, and younger workers face a 1.6 times higher likelihood than workers in their prime career years.
McKinsey sorts the 11 million transitioning workers into three types of paths. About 14 percent have a direct path, meaning they can move into a growing job with little retraining and no pay cut. Another 41 percent face a winding path, needing moderate retraining and possibly a temporary pay cut. The remaining 45 percent are on what the report calls an unpaved path: large skill gaps, and often a credential or certification requirement that takes real time to earn. In fact, roughly 85 percent of growing jobs now ask for some kind of credential, which is one of the biggest practical barriers workers actually run into.
This lines up with what we covered in our earlier look at the World Economic Forum’s four possible futures for AI and jobs by 2030: the outcome depends heavily on how much support workers get, not just on the technology itself.
What You Can Actually Do About It
You cannot control McKinsey’s projections, but you can control how ready you are. A few practical steps worth taking this month:
- Spend an hour looking honestly at which parts of your current job are repetitive and could shift to AI tools, and which parts genuinely need human judgment. The second group is where your value is growing.
- Pick one AI tool relevant to your field and actually use it on real work, not just a demo. Comfort matters more than mastery right now.
- If your field requires a credential to move up, look into free or low-cost options before paying for an expensive course. Our roadmap for learning AI for free is a good starting point.
- If you are actively job hunting, know which AI skills employers actually look for before you put them on paper. We broke that down in what AI skills on your CV are really worth.
None of this guarantees a smooth transition. But McKinsey’s own numbers show the difference between a direct path and an unpaved one usually comes down to preparation that starts months or years before you need it, not after.
Common Questions
Is this report saying AI will cause mass unemployment?
No. McKinsey actually projects a net gain of around 5 million jobs in the US by 2035. The concern is about the difficulty of moving between shrinking and growing occupations, not a shortage of jobs overall.
Which jobs are shrinking the fastest?
Office and administrative support, retail and sales, and transportation and logistics are seeing the steepest declines in demand, according to the report.
Which jobs are growing?
Healthcare support and professional roles, construction, management, manufacturing, education, and information technology are all projected to grow.
What is the single most useful skill to build right now?
McKinsey’s data points to AI fluency and adaptability as the fastest-growing in demand, and neither requires a technical degree to develop.
Final Takeaway
McKinsey’s report is one of the clearest pictures yet of what AI job transitions will actually look like this decade: not a wave of unemployment, but a much faster shuffle of who does what, with real winners and real people who get left on a harder path if they wait too long to start adapting. The most useful thing you can take from it is not the 11 million figure. It is the reminder that the workers with a direct path forward are usually the ones who started building AI fluency and adjacent skills before they needed to.
by admin | Sep 29, 2026 | AI Tools
How many browser tabs do you have open right now just to get one task done? A doc in one tab, a project board in another, maybe a CRM or a spreadsheet somewhere in the mix. Google just made a move to close a few of those tabs for good.
On September 15, 2026, Google announced new Gemini Google Workspace integrations that let Gemini reach into outside business tools like Asana and Salesforce directly from Docs, Sheets, Gmail, and Slides. No copy pasting, no switching tabs, no exporting a file just to check a number. Here’s what actually changed and who gets to use it.
What Google Actually Announced
According to Google’s official Workspace Updates blog, Gemini can now connect to seven outside platforms: Asana, Atlassian Rovo, HubSpot, Mailchimp, QuickBooks, Monday, and Salesforce. Once connected, you can ask Gemini a question about any of them while you’re sitting inside a Google Doc, a Gmail draft, or a Sheet, and it will pull the answer in without you ever opening a new tab.
The connections run on something called the Model Context Protocol, or MCP for short. In plain terms, MCP is an open standard that lets an AI model like Gemini talk to outside apps in a consistent way, instead of every company building its own one off connection. The protocol’s own documentation describes it as a common language for connecting AI systems to the tools and data they need, and it’s quickly becoming the plumbing behind a lot of the agentic AI features you’ll see rolled out this year.
What This Looks Like Day to Day
The point of these Gemini Google Workspace integrations isn’t a flashy demo, it’s the boring middle part of work that eats your morning. A few real examples:
- Ask Gemini in Gmail to check the status of a client’s project in Asana before you reply to their email
- Pull a customer’s recent activity from Salesforce or HubSpot while drafting a follow up in Docs
- Check an invoice or expense in QuickBooks without opening the accounting software
- Reference a ticket from Atlassian Rovo or a task board in Monday while writing a status update in Slides
From my own experience running websites and small online projects, this is exactly the kind of friction that quietly costs the most time. It’s rarely one big task that slows a workday down, it’s forty small ones: checking a tool, copying a number, switching back. Cutting that out, even partly, adds up fast.
Who Gets This (and Who Doesn’t Yet)
These Gemini Google Workspace integrations aren’t available to everyone yet. Per Google’s announcement, access is rolling out to Business Standard and Plus, Enterprise Standard and Plus, Google AI Pro and Ultra personal subscribers, and education plans including Google AI Pro for Education. If you’re on a free personal Google account, this one isn’t for you yet.
Admins control the feature at the domain or team level, and Google says it’s enabled by default where eligible, so if your workplace qualifies, it may already be quietly turned on. It’s worth checking your Gemini settings rather than assuming it isn’t there.
Tip: before you connect any of these tools, check with your IT or admin team about what Gemini is allowed to access. These connectors can read real business data like customer records and invoices, so it’s worth understanding your organization’s own rules before you start pulling Salesforce data into a shared Google Doc.
Why This Fits a Bigger Pattern
This isn’t an isolated Google feature. It’s part of a wider shift where AI tools stop being a separate chat window and start acting more like a coworker who can actually go check things for you. We saw a version of the same idea in how OpenAI’s researchers now lean on AI agents to handle repetitive research legwork, freeing themselves up for judgment calls. Connecting Gemini to your actual work tools is the same idea applied to everyday office work instead of a research lab.
If you want a wider look at tools like this before deciding what’s worth adopting at work or school, our roundup of useful AI tools for daily work and study is a good starting point, and our guide on how AI can help with research and productivity covers the broader habits worth building around tools like this.
Common Questions
Do I need to install anything to use these integrations?
No separate app download. If your plan qualifies and your admin has it enabled, you connect each tool once through Gemini’s settings, then ask questions about it directly inside Docs, Gmail, Sheets, or Slides.
Is my data safe when Gemini connects to a tool like Salesforce or QuickBooks?
Google says the connections are controlled through admin settings, and each organization decides what Gemini can access. Ask your own IT or admin team about your workplace’s specific data rules before connecting anything with sensitive records.
What is the Model Context Protocol, in one sentence?
It’s an open standard that lets AI models connect to outside apps and data sources in a consistent way, instead of every company building a custom connection from scratch.
Will more tools get added later?
Google hasn’t promised a specific list, but the direction is clear. MCP is being adopted widely across the industry, so more connectors are a reasonable bet over time.
Final Takeaway
The most useful AI features aren’t always the ones that generate something new, sometimes they’re the ones that just remove a step you didn’t realize was slowing you down. These Gemini Google Workspace integrations won’t change what you do at work, but they might change how many tabs you need open to do it. If your plan qualifies, it’s worth five minutes to check your Gemini settings and see what’s already connected.
by admin | Sep 27, 2026 | AI Guides
If you have noticed more headlines about AI and cybersecurity lately, you are not imagining it. In the space of about a month, Google, Anthropic, and OpenAI each put out a major announcement tying their newest models directly to cyber defense. Three different companies, three different products, one clear signal: the AI labs think cybersecurity is where their models are about to matter most.
You do not need to run a security team to care about this. These announcements affect how safe your data is, how fast the next big vulnerability gets patched, and eventually, what AI tools look like when you use them at work. Here is what actually happened with each of these AI cybersecurity models, in plain English.
Google’s Fairwind Program: Patching Holes Before Attackers Find Them
On September 2, 2026, Google launched the Fairwind Program, a limited access setup that pairs its new Gemini 3.8 Flash Cyber model with a tool called CodeMender. Together they let security teams find a vulnerability and generate a verified, ready to deploy patch in minutes instead of the weeks that normally takes.
Access is not open to everyone. Google built this for governments, national cyber authorities, and operators of critical infrastructure like hospitals, power grids, and telecom networks, plus Google Cloud customers and security partners. The company says more than 650 organizations are already part of the program, including names like CrowdStrike, Palo Alto Networks, and Snowflake. The logic is simple: give the people defending hospitals and power plants a head start before the next major exploit shows up.
Anthropic’s Enterprise Frontier Safeguards: Watching for Misuse Without Holding Your Data
Anthropic announced its own piece the day before, on September 1. Called Enterprise Frontier Safeguards, it tries to solve a real problem for businesses: to catch a slow, sneaky attack (one that unfolds over several sessions instead of a single obvious moment) you need to keep some activity data around. But a lot of companies, especially in regulated industries, do not want Anthropic holding that data at all.
The fix is that the monitoring data lives in the customer’s own cloud storage, whether that is Amazon S3, Azure, or Google Cloud, not on Anthropic’s servers. Automated systems still flag suspicious patterns and send alerts, but no Anthropic employee reviews the underlying data unless the customer wants that. Anthropic says the rollout starts in phases this fall, with zero data retention already available on its Fable 5 and 5.1 models in the meantime.
OpenAI’s Astra: The First Model Rated “Critical” for Cyber Capability
The biggest claim of the three came from OpenAI. Under its own Preparedness Framework, the company says its Astra model is the first it has ever rated at the “Critical” level for cybersecurity capability, meaning it could, in theory, find and use a working zero day exploit against a well defended system without a human guiding every step.
That is exactly why OpenAI paired the announcement with a long list of safeguards rather than a wide release. Astra refuses 91.5% of attempts to trick it into cyber misuse, compared to 59% for its predecessor GPT-5.6 Sol, and in testing it made zero attempts to interfere with the security systems watching it, versus 56% for the older model. For now, the advanced cyber capabilities are only going to a small group of alpha testers through something called the Daybreak Blue program, with wider access planned later for defensive use only.
Quick tip: none of these tools are available to the general public right now. If an email, ad, or “early access” link claims to offer you Gemini Cyber, Astra, or Claude’s cyber features directly, treat it as a scam. Every one of these programs is invite only and vetted.
What These AI Cybersecurity Models Mean for You
From my own experience working with websites, online tools, and cybersecurity, the pattern here is familiar. Attackers have always moved fast, and defenders have always been playing catch up. What is new is that the companies building the most capable AI models are now openly admitting those models are powerful enough to matter on both sides of that fight, and building the guardrails in public instead of quietly.
- Run a small business or a website: faster patching tools like Fairwind mean fewer known vulnerabilities sitting unfixed, which helps even if you never touch the tool yourself.
- Use Claude or ChatGPT at work: safeguards like Enterprise Frontier Safeguards and Astra’s refusal training are part of why IT teams are more willing to approve AI tools for everyday use.
- Learning AI or eyeing a tech career: this is a strong signal that cybersecurity and AI safety work is becoming one of the more stable places to build one, a trend covered in our piece on why cybersecurity careers are booming in the AI era.
It is worth remembering that more capable models cut both ways. We covered the other side of this story, the rise in AI assisted cyberattacks, in our breakdown of Anthropic’s report on AI cyberattacks. These September announcements are the labs’ answer to that same trend, and if you already use Claude at work, they are landing alongside product news too, including the recent Claude for Small Business upgrade.
Common Questions
Can I use Gemini Cyber, Astra, or Claude’s cyber features right now?
No. All three programs (Fairwind, Enterprise Frontier Safeguards, and Daybreak Blue) are limited to approved organizations and testers, not the general public.
Does this mean AI can now hack anything on its own?
Not quite. OpenAI’s “Critical” rating for Astra means the model has crossed a capability threshold that requires stronger safeguards, not that it can bypass any system unsupervised. That is exactly why access is restricted and monitored.
Is my data safer because of Anthropic’s Enterprise Frontier Safeguards?
If your employer uses Claude for business and adopts EFS, your company’s activity data stays in your own company’s cloud storage rather than Anthropic’s servers, while still being monitored for misuse.
Why does this matter if I am not in tech?
Faster patching and stronger AI safeguards reduce the number of unfixed vulnerabilities across the systems you use every day, from your bank’s website to your employer’s internal tools, even if you never interact with these models directly.
If you want to understand the basics of AI before topics like this start to make sense, our simple explanation of what AI is is a good place to start.
Final Takeaway
Three of the biggest AI labs spent one month proving that AI cybersecurity models are no longer a side project. Google is speeding up patching, Anthropic is rethinking how monitoring data is stored, and OpenAI is being unusually open about a model crossing into genuinely risky territory. None of it changes your day tomorrow. But it is a good reminder to keep your own software updated, question anything that claims to offer you “early access” to these tools, and pay attention, because the gap between what attackers can do and what defenders can do is exactly what these companies are racing to close.
