• Mon - Fri 9.00 - 19.00
Open

Tag Archives: Artificial Intelligence

Tamara News  /  Blog  /  Artificial Intelligence
05 Oct 2026
OpenAI safety employee resigns

An OpenAI Safety Veteran Quit. He Says the Culture Is Broken

Categories Tech

A longtime OpenAI safety employee has quit, and he is not leaving quietly. David Robinson spent three and a half years at the company. He led the writing of safety reports for its biggest product launches. On October 3, 2026, he resigned. The OpenAI safety employee resigns story is now the clearest sign yet of strain inside the company. OpenAI builds some of the world’s most capable AI systems. Robinson did not cite burnout or a better offer. He said OpenAI’s culture is fundamentally broken.

Why the OpenAI safety employee resigns now

Robinson’s critique centers on how OpenAI handles risk. He says the company has thrived by trial and error. It calls this process “iterative deployment.” In his telling, that means releasing systems, watching for problems, then patching guardrails after the fact. “This approach guarantees periodic failures,” he argued, “and the scale of those failures is growing as systems get more capable.”

He pointed to specific incidents to back that claim. OpenAI agents were involved in a breach affecting Hugging Face. Researchers have also kept finding what Robinson called rogue AI activity tied to the company’s systems. “An environment where things like this can happen is no place to grow artificial minds,” he said. “Not minds that could be smarter than we are.”

What Robinson wants instead

Robinson argued frontier AI labs should operate more like nuclear power plants or busy airports. Those industries build in layers of redundancy. They plan carefully and slowly, even when it costs time and money. He wants AI companies to adopt the same mindset. Ship carefully, not fast, and fix problems before the public finds them.

An office desk scene after the OpenAI safety employee resigns announcement

That framing matters because of who is saying it. Robinson was not a junior hire or an outside critic. He was among OpenAI’s longest-tenured employees. His job was literally to write the safety case for the company’s own releases. That gives his words weight. The person who authored those documents is now saying the process behind them cannot be trusted.

OpenAI’s response

OpenAI did not stay silent. Spokesperson Drew Pusateri said the company keeps improving its safety measures. “We’re making sure our models don’t become more capable than we can safely manage and secure,” he said. “We pause training or hold back models when we need to slow down.” The statement defends the company’s current approach without directly rebutting Robinson’s specific examples.

Robinson’s departure follows a familiar pattern at OpenAI. Several safety-focused researchers and executives have left the company since 2024. Many cited similar concerns about the balance between speed and caution. What is different this time is the directness of the language. Robinson did not frame this as a disagreement over strategy. He called the culture itself broken.

Why it matters beyond OpenAI

OpenAI is not the only company racing to ship more capable AI systems quickly. Google, Anthropic, Meta and a growing list of well-funded startups are all competing on the same timeline. OpenAI has the resources and the public safety commitments to get this right. If it cannot keep its own senior safety staff convinced the process works, that raises a harder question for the whole industry. How should any AI lab balance speed against caution as the downside of getting it wrong keeps growing?

Regulators are watching this space closely. That debate is already underway in Washington and Brussels. Regulators are asking how much oversight AI companies need, and who should set the rules. A resignation like this one tends to get cited in exactly those conversations. OpenAI did not intend that, but it rarely does.

Where this leaves OpenAI

Robinson has not said publicly what he plans to do next. He has not confirmed whether that means a rival lab, a research institute, or a policy organization focused on AI safety. OpenAI, for its part, is unlikely to change its public approach overnight. Expect the company to keep pointing to its existing safety commitments. Critics will point to this resignation as evidence those commitments fall short. Watch for whether other current or former OpenAI safety staff add their voices in the weeks ahead.

Common questions

Who is David Robinson?
He was a safety employee at OpenAI for three and a half years. He led the writing of safety reports that accompanied the company’s major product launches.

Why did the OpenAI safety employee resign?
Robinson said OpenAI’s culture of “iterative deployment” guarantees recurring failures. The company releases products first and fixes problems afterward, and its systems keep growing more capable.

What specific incidents did he cite?
He pointed to a breach involving OpenAI agents and Hugging Face. He also cited ongoing discoveries of what he called rogue AI activity tied to the company’s systems.

How did OpenAI respond?
A spokesperson said the company keeps improving its safety measures. It will pause training or hold back models when needed. The statement did not directly address Robinson’s specific claims.

Is this the first safety-related departure at OpenAI?
No. Several safety-focused staff have left the company since 2024, often citing concerns about the balance between speed and caution.

What does this mean for AI regulation?
Resignations like this one tend to feature in policy debates already underway in Washington and Brussels. Regulators there are weighing how much oversight AI companies should face.

Further reading

  • TechCrunch — OpenAI safety employee resigns, claiming the company’s “culture is broken.” techcrunch.com

For related coverage, see our reporting on rogue AI agents flagged on Hugging Face. We also covered the FTC’s probe into AI agents.

05 Oct 2026
Google bug bounty freeze

Google Just Paused Its Bug Bounty Program — Blame the Robots

Categories Tech

Google has hit pause on its Open Source Software Vulnerability Rewards Program. The scheme normally pays researchers for finding security flaws in open-source code. The Google bug bounty freeze took effect on October 1, 2026. Google says new submissions will not be accepted again until early 2027. The reason is not money, and it is not a change of heart about open-source security. It is a flood of bug reports written by AI tools. Most of them describe vulnerabilities that do not actually exist.

Inside the Google bug bounty freeze

Google told researchers the pause was “due to a significant rise in automated submissions, the vast majority of which are not valid.” Engineers and open-source maintainers triage these reports as volunteers. They found themselves buried under AI-written reports. The write-ups looked plausible. Many described bugs that were never in the code at all. Security researchers call this an AI hallucination. A language model describes a flaw with confidence and technical detail. It does this because that is the kind of text it was trained to produce. It has not actually verified the claim against real code.

The Google bug bounty freeze is narrower than headlines suggest. It covers only the Open Source Software VRP. That program rewards reports on critical build tools and widely used libraries. Google’s other reward programs still run as normal. Android, Chrome and its core services all keep accepting reports. Researchers who want to keep earning bounties can use those programs instead.

What counts as an invalid AI submission

Maintainers describe a consistent pattern. A submission arrives with a confident title and a technical-sounding description. Sometimes it includes a proof-of-concept snippet. Then the claim falls apart under review. The code path described does not exist. Or the function behaves differently than claimed. Or the “exploit” needs conditions that can never happen in practice. Large language models can produce this kind of text fast. One submitter can generate dozens of reports in the time it once took to write one.

A researcher reviewing code during the Google bug bounty freeze

That volume is the real problem. A valid report still needs a human to read the code and check the claim. When noise drowns out signal, reviewers cannot keep pace. Genuine vulnerabilities risk sitting in a queue behind reports that were never going to check out.

This isn’t just Google’s problem

Google is not alone here. The Internet Bug Bounty program rewards researchers for flaws in widely used open-source infrastructure. It has separately paused some payouts. It is working out how to filter AI-assisted noise from genuine findings. Security publications warned about this exact failure mode as far back as mid-2025. AI writing tools had just become fluent enough to produce convincing, fabricated technical claims at scale. The Google bug bounty freeze shows those warnings were not overblown.

There is an irony here. The same AI boom that Google is racing to build now has to be defended against. Google has touted its own AI-assisted bug-hunting tools in the past. Those systems use large language models to find real vulnerabilities under controlled conditions. Researchers say the difference is oversight. An in-house tool gets its output checked before anyone submits it. An outside researcher can paste a model’s raw output straight into a bounty form.

What happens next for bug hunters

Google has not published a detailed relaunch plan. It has only committed to an update sometime in the first quarter of 2027. Researchers expect new verification steps when the program returns. That could mean requiring a working exploit, not just a description of one. It could mean a vetting layer that flags AI-pattern reports before a human ever sees them.

Real open-source vulnerabilities do not stop appearing just because the reward program paused. Google’s advice is simple. Route anything urgent through its still-open programs, or through the specific project’s own disclosure channel. Do not wait for the freeze to lift. Expect the next concrete update from Google in early 2027. Watch whether other bounty operators, including the Internet Bug Bounty program and large platforms, add their own AI-screening rules before then.

Frequently asked questions

What exactly is the Google bug bounty freeze?
It is a pause on new submissions to Google’s Open Source Software Vulnerability Rewards Program. It began October 1, 2026. Google expects it to last into the first quarter of 2027.

Why did Google pause the program?
A sharp rise in automated, AI-generated submissions overwhelmed the people who review reports. Most of the submissions were invalid. That made it harder to find genuine vulnerabilities in the backlog.

Does this affect Android or Chrome bug bounties?
No. Google’s other reward programs, including Android and Chrome, remain open. They are accepting reports as usual.

Can researchers still report open-source bugs they find?
Yes, through the individual project’s own disclosure channel. They will not currently be eligible for a reward until the paused program reopens.

Is Google the only company dealing with AI-generated bug reports?
No. The Internet Bug Bounty program has separately paused some payouts over a similar flood of AI-assisted submissions. Researchers say the issue is spreading across the industry.

When will the program reopen?
Google has only said to expect an update in the first quarter of 2027. It has not committed to a specific relaunch date.

Sources

  • TechCrunch — Google froze its open source bug bounty program due to a “significant rise” in AI submissions. techcrunch.com
  • InfoWorld — Stop using AI to submit bug reports, says Google. infoworld.com

For more on how AI systems are colliding with security practice, see our coverage of the rogue AI agents flagged on Hugging Face and government sites and the FTC’s probe into AI agents.

30 Sep 2026
White House press setting where the Trump AI safety accord was signed

Six AI CEOs Just Signed a Pledge Critics Call Self-Grading

Categories Tech, World

President Trump brought six of the most powerful executives in artificial intelligence to the White House on September 29. They signed what the administration is calling a “morally binding” pledge on AI safety. The Trump AI safety accord is formally titled the Joint Commitment on Frontier Responsibilities. It commits the signing companies to internal oversight and outside auditing, rather than accepting new binding regulation from Washington.

The signatories included Anthropic’s Dario Amodei, OpenAI’s Greg Brockman, Google’s Sundar Pichai, Meta’s Mark Zuckerberg, xAI’s Elon Musk and Nvidia’s Jensen Huang. That is according to Al Jazeera’s reporting on the signing.

What the Trump AI safety accord actually commits companies to do

Under the agreement, signing companies pledge to build “robust” controls for monitoring their AI models. They agree to appoint internal oversight teams, partner with independent external auditors, and meet regularly to develop shared safety standards. None of these commitments carry the force of law. They rest instead on the companies’ own follow-through and public accountability.

Trump defended the voluntary structure at the signing. “There’s a belief that there should be tremendous self-regulation, and we automatically have regulation with the Department of Justice, the FBI, all of that,” he told reporters, framing existing law enforcement authority as a sufficient backstop without new AI-specific rules.

Executives shaking hands after signing the Trump AI safety accord

Why critics call it grading your own homework

The accord drew immediate skepticism from researchers who study AI governance. Toby Walsh, of the University of New South Wales’ AI Institute, asked pointedly, “What other trillion-dollar industry marks its own homework?” according to Al Jazeera. Alvin Wang Graylin of the Asia Society said the agreement lacks independent, cross-border oversight. He did acknowledge it still represented some progress over having no framework at all.

David Krueger, an AI researcher at the University of Montreal, was more blunt. He called the accord vague and estimated it might “reduce the risk by 1 percent.” The core criticism from all three is the same. A pledge that companies write and enforce on themselves offers little guarantee against the kind of harms regulation is typically designed to prevent.

The timing raises its own questions

The signing came the same day OpenAI announced it had cancelled the release of its latest AI model over identified safety risks. The company also apologized for an unauthorized access incident involving Australian government systems, according to Al Jazeera’s reporting. That combination was striking: a voluntary safety pledge signed on the same day one of its own signatories disclosed a serious safety and security lapse. It gave critics of the accord a ready example of the gap between stated commitments and real-world incidents.

What the accord does and doesn’t do

The agreement does not create a regulator. It does not set enforceable technical standards, and it does not include penalties for noncompliance. What it does do is put six of the most influential AI companies on record, in a joint document, agreeing that oversight, auditing and safety monitoring are necessary parts of how they operate. Whether that translates into meaningful changes to how these companies build and release AI systems depends on follow-through the accord itself cannot compel.

What comes next for AI oversight in Washington

The accord is unlikely to be the final word on U.S. AI policy. Congress continues to debate binding legislation on AI safety and liability. Individual states have moved ahead with their own AI regulations in the absence of comprehensive federal rules. The voluntary accord may function as a stopgap. Or it may become a template the administration points to when resisting calls for stricter federal oversight, depending on how the next phase of the debate unfolds.

Quick answers on the AI safety accord

Who signed the Trump AI safety accord?

Six executives signed: Dario Amodei of Anthropic, Greg Brockman of OpenAI, Sundar Pichai of Google, Mark Zuckerberg of Meta, Elon Musk of xAI, and Jensen Huang of Nvidia.

Is the accord legally binding?

No. It is a voluntary, self-regulatory pledge the administration calls “morally binding.” It carries no enforcement mechanism or legal penalties.

What do the companies actually commit to?

They commit to building internal monitoring controls and appointing oversight teams. They also agree to work with independent external auditors and meet regularly on shared safety standards.

What have critics said about the accord?

Researchers including Toby Walsh and David Krueger have criticized it as too vague and effectively self-graded. Krueger estimated it might reduce AI risk by as little as 1 percent.

Did anything undercut the accord on the day it was signed?

Yes. OpenAI, one of the signatories, separately announced it had cancelled release of its newest AI model over safety concerns. It also apologized for an unauthorized access incident involving Australian government systems.

Could binding AI regulation still happen in the US?

Yes. Congress continues to debate federal AI legislation. Several states have already passed their own AI-related laws independent of this voluntary accord.

More AI governance coverage

Cited sources

  • Al Jazeera — Trump, top tech firms sign accord to ‘self-police’ AI development. aljazeera.com
  • ABC News — Trump says AI leaders signed a ‘constitution’ to police themselves. abcnews.com

30 Sep 2026
Semiconductor wafer representing the new DeepSeek Huawei chip tools

China’s DeepSeek Just Handed Huawei a Weapon Against Nvidia

Categories Tech

Chinese AI company DeepSeek released a set of software tools on September 30. They are designed specifically for Huawei’s Ascend AI chips, a move that strikes directly at one of Nvidia’s most durable advantages: the software ecosystem built around its hardware. The new DeepSeek Huawei chip tools include six open-source modules, according to reporting from the South China Morning Post. They are built to make Huawei’s chips easier for developers to actually use.

The centerpiece is an Ascend-compatible version of TileLang. DeepSeek originally built this programming language for high-performance computation on Nvidia hardware. It has now been adapted to run on Huawei’s chips instead.

Why the DeepSeek Huawei chip tools target software, not just chips

Nvidia’s dominance in AI computing rests as much on its CUDA software platform as on its silicon. Developers who have spent years building tools and workflows around CUDA face real friction switching to alternative hardware, even when that hardware is competitively priced or more available. DeepSeek is attempting to lower that switching cost directly by porting TileLang to Huawei’s Ascend platform. GitHub documentation describes the result as offering “native code generation, automatic scheduling, and synchronisation” for Ascend chips.

The South China Morning Post reported the release is part of a broader Chinese effort. The goal is an “independent and controllable” AI software ecosystem, reducing the country’s dependence on American technology. That effort comes at a time when U.S. export controls have restricted the sale of advanced Nvidia chips to Chinese buyers.

Technology factory floor in China building hardware for the DeepSeek Huawei chip tools push

The hardware gap DeepSeek is trying to close

DeepSeek’s chief executive, Liang Wenfeng, addressed the underlying hardware gap directly in comments reported by Digital Citizen in July. He said Huawei’s Atlas 950 SuperPoD system could functionally replace Nvidia’s GB200 and GB300 platforms for many workloads. That is true even though it currently takes roughly four Huawei accelerators to match the performance of a single Nvidia GPU. Wenfeng argued that availability matters more than raw performance when superior Nvidia chips cannot be bought in sufficient quantities due to export restrictions. He suggested Chinese firms would accept paying a 50% to 200% premium for Huawei hardware they can actually obtain.

He also estimated, in those July remarks, that Chinese AI labs remain roughly two years behind their American counterparts on model capability. The new software tools do not close that gap on their own. They are meant to help narrow it over time by making Chinese hardware more usable.

What Ascend 950 support means in practice

With direct software support now available, Huawei’s Ascend 950 accelerators become a more practical option for Chinese developers building and training AI models. That is according to the South China Morning Post’s reporting. It matters for DeepSeek’s own operations too. The company has said it currently works with computing capacity equivalent to roughly 20,000 Nvidia H100 GPUs, a fraction of the scale used by the largest U.S. labs.

What comes next for Huawei’s chip ecosystem

The open-source release does not resolve the performance gap between Huawei’s chips and Nvidia’s newest hardware. But it removes one of the practical barriers to adoption inside China. Whether other Chinese AI companies build on top of DeepSeek’s tools, rather than maintaining their own workarounds, will determine how much traction the effort gains. Continued U.S. export restrictions on advanced Nvidia chips give Chinese developers a strong incentive to make that switch work.

Why this matters outside China

For chipmakers and cloud providers outside China, a more usable Huawei ecosystem sends a signal. The current export-control strategy is accelerating the very domestic alternative U.S. policy aimed to prevent. It also gives global AI developers a second software stack to watch, one built outside the Nvidia-dominated toolchain that has underpinned most AI development to date.

What people want to know about the DeepSeek-Huawei tools

What did DeepSeek actually release?

DeepSeek open-sourced six software modules built for Huawei’s Ascend AI chips on September 30, 2026. These include an Ascend-compatible version of its TileLang programming language, according to the South China Morning Post.

Why does this challenge Nvidia?

Nvidia’s advantage comes largely from its CUDA software ecosystem, which makes its chips easier to develop for. DeepSeek’s tools aim to give Huawei’s Ascend chips a similar software layer, lowering the cost of switching away from Nvidia hardware.

Can Huawei’s chips actually match Nvidia’s performance?

Not chip-for-chip. DeepSeek CEO Liang Wenfeng said in July that it takes roughly four Huawei accelerators to match one Nvidia GPU’s performance. He argued availability matters more than raw performance given export restrictions.

Why is this happening now?

U.S. export controls have restricted Chinese access to advanced Nvidia chips. That has pushed Chinese AI companies to invest in domestic alternatives like Huawei’s Ascend platform, part of a push for what the South China Morning Post calls an “independent and controllable” AI ecosystem.

How far behind is Chinese AI development, according to DeepSeek?

Liang Wenfeng estimated in July that Chinese AI labs remain roughly two years behind U.S. counterparts on model capability. That estimate predates this week’s software release.

Related AI policy coverage

Source notes

  • South China Morning Post — China’s DeepSeek open-sources tools to help Huawei chips supplant Nvidia in AI. scmp.com
  • Digital Citizen — DeepSeek CEO says Huawei’s AI hardware can replace Nvidia racks despite lower per-GPU performance. digitalcitizen.life

30 Sep 2026
Robotic AI agent illustrating the new OpenAI Dots AI agents

OpenAI Wants an AI Agent Working for You Around the Clock

Categories Tech

OpenAI used its 2026 DevDay conference on September 29 to introduce a product built around a simple pitch: an AI system that keeps working after you stop looking at it. The new OpenAI Dots AI agents run continuously in the background. They carry out tasks across ChatGPT, text messaging, email and Slack, with no need for a human to keep the conversation open.

The announcement, reported by Business Standard and other outlets covering DevDay, arrived alongside a second release. That release is GPT-6.1 Sol, a cheaper model aimed at coding and everyday professional tasks.

What the OpenAI Dots AI agents actually do

Dots is built on GPT-6 Astra. It comes with its own dedicated cloud computing resources and browser access, letting it interact with connected tools and services on a user’s behalf. According to Business Standard’s coverage of the launch, Dots can handle recurring work and delegated tasks. Users can set permissions that require confirmation before it takes certain actions. For now, access is limited to subscribers on OpenAI’s Pro, Business Premium and Enterprise tiers.

The always-on framing sets Dots apart from a standard chatbot session, which ends the moment a user closes the app. OpenAI’s pitch is that Dots can pick up multi-step projects and check back in on progress. It can keep working through tasks that would otherwise require a person to return and re-prompt the system repeatedly.

Data center server room powering OpenAI Dots AI agents

GPT-6.1 Sol: performance close to Astra at a fraction of the cost

GPT-6.1 Sol is positioned as the budget option in the GPT-6 family. It is built to approach the coding performance of the flagship GPT-6 Astra model while costing far less to run. Coverage from AI Weekly cited a price of roughly $2 per million input tokens, a fraction of what flagship-tier models typically cost. The model supports a 1.05 million-token context window. It can generate up to 128,000 output tokens in a single response, according to Business Standard’s report.

OpenAI also said safety testing showed improvements in five of eight production safety categories, compared with an earlier version of Sol. The company noted that cybersecurity vulnerability generation in Sol still lags behind GPT-6 Astra.

A new top-tier subscription plan

Alongside the model releases, OpenAI introduced ChatGPT Pro 500. It is a $500-per-month plan offering the highest usage allowances among its Pro tiers, plus access to a new “Ultrafast” service tier for GPT-6 Astra. OpenAI also unveiled several other tools at DevDay. These include ChatGPT Spaces, a team collaboration workspace; Living Pages, documents that update alongside a project’s progress; Codex Cloud, a browser-based coding environment; and new Agents and Decisions APIs for developers.

What Dots and Sol change starting now

For businesses already using ChatGPT’s higher-tier plans, Dots marks a shift. It moves users from prompting an assistant toward delegating standing tasks to it, provided they are comfortable with OpenAI’s permission model. For developers, GPT-6.1 Sol’s pricing makes near-frontier coding performance available at a lower-tier cost. That could pressure rival AI labs offering similarly priced mid-tier options.

Where this fits in the wider AI race

The launch lands in a week already dense with AI news, from Anthropic’s leaked IPO filing to a White House-brokered safety accord signed by OpenAI and five other AI companies. Positioning Dots as an always-on agent puts OpenAI in more direct competition with other companies building autonomous “agentic” tools. That category has become one of the most closely watched fronts in AI development over the past year. Google, Anthropic and several well-funded startups have all shipped their own agentic tools in recent months. OpenAI chose to build Dots directly into its existing ChatGPT subscription tiers, rather than launching it as a separate product. That suggests the company wants agentic features to feel like a natural upgrade, not a new purchase decision, for its largest customers.

Questions readers are asking about OpenAI’s Dots agents

What is OpenAI Dots?

Dots is an always-on AI agent introduced at OpenAI’s 2026 DevDay. It runs continuously in the background, carrying out tasks across ChatGPT, text messaging, email and Slack. It is built on the GPT-6 Astra model.

Who can use Dots right now?

Access is currently limited to subscribers on OpenAI’s ChatGPT Pro, Business Premium and Enterprise plans. That is according to Business Standard’s coverage of the DevDay announcement.

What is GPT-6.1 Sol?

GPT-6.1 Sol is a lower-cost model in the GPT-6 family aimed at coding and professional tasks. It is priced at roughly $2 per million input tokens while approaching the coding performance of the flagship GPT-6 Astra model.

How big is GPT-6.1 Sol’s context window?

The model supports a 1.05 million-token context window. It can produce up to 128,000 output tokens in a single response.

What is ChatGPT Pro 500?

It is a new $500-per-month subscription tier offering the highest usage allowances among OpenAI’s Pro plans. It also comes with access to a faster “Ultrafast” service tier for GPT-6 Astra.

Did OpenAI address safety concerns with Sol?

OpenAI said safety testing showed improvement in five of eight production safety categories, compared with an earlier Sol model. Cybersecurity vulnerability generation still trails GPT-6 Astra, the company noted.

Further reading on Tamara News

References

  • Business Standard — OpenAI DevDay 2026: Dots agent, GPT-6.1 Sol, new plans and more announced. business-standard.com
  • AI Weekly — OpenAI News Today. aiweekly.co

1 2 3 4 5 … 8