Hasty Briefsbeta

All tags

#ai ethics

39 stories total

Bilingual

Weird Al declined 'a nice pile of money' to star in AI ad
22 days ago
- "Weird Al" Yankovic declined a lucrative commercial offer due to his opposition to AI.
- He learned the product was AI a week before the shoot and withdrew to avoid being its "poster boy."
- Yankovic publicly stated he is "not a fan of AI" and prioritized integrity over money.
Godot will no longer accept AI-authored code contributions
23 days ago
- Godot will amend contribution guidelines to reject AI-authored code and pull requests
- AI-generated contributions have become draining for maintainers and hinder mentorship of new contributors
- The policy requires human accountability for code and rejects AI text in human communication
- Contributors must disclose any AI assistance used for menial tasks only
- Godot maintains a conservative approach but will re-evaluate as AI tools evolve
Political bias in AI: Where the AI models stand
a month ago
- AI models are evaluated on political bias using repeated questions without web search.
- Models are plotted as clouds showing their range of answers on political and economic axes.
- Each model's position is compared to real-world figures from expert surveys.
- A gap is noted between what models claim about their leanings and where they actually measure.
- The project emphasizes descriptive analysis without partisan judgment.
- Methodology includes open question bank, run-to-run stability, and refusal tracking.
Open letter to Anthropic: keep Claude Fable 5 in existing paid plans
21 days ago
- Anthropic gave all paid subscribers access to Claude Fable 5, the most advanced AI system, from July 1st.
- On July 7th, they plan to remove Fable 5 from subscription plans, requiring users to purchase additional usage credits to retain access.
- The letter argues that pricing is not a safety control, as Fable 5 is already deemed safe for public use.
- Access to advanced AI is essential for economic activities like job applications, small businesses, and research, and usage-based pricing risks creating economic disparities.
- Anthropic, as a public benefit corporation, is called on to ensure AI benefits are broadly shared, not restricted by income.
- The promotion proved Fable 5 could be included in subscription plans within usage limits, showing feasibility.
- The undersigned demand Fable 5 remain in paid plans, a published diffusion policy, and a commitment that price is not used as a safety measure.
AI agents are not your "coworkers"
20 days ago
- Marketing AI agents as digital employees can reduce human error detection by 18% and shift accountability.
- Framing AI as coworkers leads to less human responsibility and more reliance on managerial oversight, negating efficiency gains.
- AI agents should enhance human capabilities rather than replace them, as suggested by Nobel-winning economist Daron Acemoglu.
- Workers prefer AI assistance in specific areas, such as tracking case progress, not in tasks like credit verification where human judgment is valued.
- Labeling AI tools as employees is a branding tactic that can degrade human performance and obscure real accountability in failures.
Godot bans "vibe-coded" code contributions
20 days ago
- Godot is updating its contribution policy to ban almost all AI-generated code contributions.
- New contributors (with ≤3 merged PRs) must get explicit permission to submit major changes.
- AI assistance is limited to menial tasks like code completion and must be disclosed.
- The team cites demoralizing, low-quality AI PRs and lack of understanding from contributors.
- Contribution discussions must remain human-to-human, with no AI agents in communications.
- Context is highlighted as crucial in software development, which AI often fails to grasp.
The Reports of Jim Carrey's Death Are a Failure Mode
20 days ago
- Google incorrectly displayed a death notice for Jim Carrey on June 29, 2026, highlighting a failure in its knowledge system.
- Knowledge systems must balance trust and verification, as information becomes knowledge only after surviving scrutiny.
- Misinformation spreads faster than verification, especially with AI systems that can amplify false claims.
- Provenance and transparency are crucial; systems should track a claim's source, conflicts, and state over time.
- Verification should be proportional to consequence, with high-risk claims requiring stricter checks.
- Systems need to represent epistemic states like uncertainty or contradiction, not just true/false binary.
- Claims should have a lifecycle (asserted, supported, disputed, retracted) to reflect real-world changes.
- AI systems can obscure uncertainty by generating confident summaries from unverified sources.
- Trustworthy knowledge systems will be those that show provenance and allow correction, not just propagate information.
Tripadvisor AI summaries give glowing reviews to dangerous hotels
18 days ago
- Tripadvisor's AI summaries are downplaying serious issues like food poisoning, sexual harassment, and hygiene failures at hotels.
- Investigation reveals AI summaries for hotels such as Riu Palace Santa Maria describe them positively despite numerous reports of illness and poor conditions.
- Guests at Riu Palace reported raw chicken, flies in food, and food poisoning, with legal action involving over 412 holidaymakers and seven deaths since 2023.
- Tripadvisor's AI tool Ollie incorrectly stated food poisoning was 'quite unlikely' at Riu Palace, contradicting guest reviews.
- AI summaries for other hotels like Garza Blanca and Occidental Caribe also omitted serious health and safety concerns from recent reviews.
- At Kaia Coracesium in Turkey, AI summarized service as 'friendly' despite multiple reports of sexual harassment from staff.
- Tripadvisor claims its AI summaries aim for transparency and are based on recent reviews, but Which? found inconsistencies in highlighting negative feedback.
- Google's AI summaries for the same hotels more accurately warn of issues like 'potential for illness' compared to Tripadvisor's.
- Which? advises travelers to scroll past AI summaries and check real guest reviews, especially one-star ratings, for safety information.
The Private Capture of Public Genius
18 days ago
- AT&T, as a regulated monopoly, funded Bell Labs, enabling decades of groundbreaking innovations.
- The 1956 antitrust settlement forced AT&T to open its patents royalty-free, catalyzing industries like semiconductors and Silicon Valley.
- AI frontier labs (like OpenAI and Anthropic) train models on vast internet data, claiming fair use and transformative value.
- Legal battles around AI training data focus on fair use, market harm, and the lack of retroactive consent from content creators.
- The internet's value as a corpus is a public good, but AI harvesting threatens participation and integrity of its layers.
- Current models cannot attribute individual contributions, making proportional payment impossible, yet value extraction is undeniable.
- A Corpus Royalty—a fixed share of AI lab revenues paid to the public—is proposed as restitution for unjust enrichment.
- Historical precedents like Alaska Permanent Fund and Superfund show collective solutions for shared resource exploitation.
- The royalty addresses unattributable creativity, ensuring the public shares in the upside of AI's reliance on collective genius.
Anthropic's Method to Losing Goodwill in a Few Easy Steps
18 days ago
- Anthropic's Claude subscription and API face issues including enshittification, vendor lock-in, consumer malpractice, and price gouging, while open-source models become more competitive.
- Claude's API is notoriously unreliable, and Claude subscriptions force users to rely on Anthropic's servers, with no compatibility with third-party harnesses without extra costs.
- Anthropic introduced 'extra usage' billing for third-party tools and split subscription billing into separate pools for first-party and third-party usage, significantly increasing costs for heavy agent users.
- The author criticizes 'vibecoding' and agent-driven development for causing skill degradation, advocating instead for agent-assisted development using AI as an autocomplete tool.
- Open-source models like Qwen, GLM, and Deepseek offer viable alternatives to proprietary models, with AI gateways enabling flexible, cost-effective, and private model usage.
- The article concludes by condemning Anthropic's unethical practices, such as locking customers into closed systems and changing terms without notice, and encourages moving to open-source alternatives.
Google Chrome Installed a 4GB AI Model on Your PC
17 days ago
- Google Chrome secretly installed a 4GB AI model (Gemini Nano) on users' devices without consent.
- The model reinstalls automatically if deleted, with an optional toggle not yet available to all users.
- Google justifies it for privacy via on-device features like scam detection, but the installation violates consent laws like the EU's ePrivacy Directive.
- Visible AI features in Chrome use cloud processing, while the local model runs background tasks, creating a 'privacy illusion.'
- This reflects a trend where tech companies use user devices as distributed AI infrastructure, raising concerns about consent and resource use.
Anthropic hid a tracker in Claude Code to flag Chinese users
17 days ago
- Anthropic secretly inserted a tracker in Claude Code to monitor Chinese users, using prompt steganography to hide code that flagged timezone, proxy, and potential connections to Chinese AI labs.
- The tracker, added as an 'experiment' in March, was intended to prevent account abuse from unauthorized resellers and protect against distillation attacks, with resellers offering cheap subscriptions.
- Anthropic quickly removed the tracker after exposure by a security researcher, who condemned it as a 'serious breach of user trust,' while privacy advocates criticized the surveillance despite Anthropic previously refusing US government surveillance requests.
Meta's AI 'Perv Glasses' Now Come with Stupid Comcast-Esque Usage Restrictions
17 days ago
- Meta has introduced rate limits on its AI smart glasses, requiring a $19.99 monthly subscription for unlimited use of the 'conversation focus' feature, which amplifies voices in noisy environments.
- The 'conversation focus' feature operates entirely on-device without needing internet or Meta's servers, making the rate limits technically unnecessary and likened to Comcast's usage caps.
- Privacy concerns have arisen due to reports of the glasses being used to video people without consent and potential facial recognition capabilities, leading to terms like 'perv glasses.'
- Critics argue Meta's move is an example of 'enshittification,' imposing arbitrary fees to boost earnings, similar to subscription models in other industries like automotive.
- The article highlights broader issues with Meta's AI investments and data center expansions, suggesting a bubble and debt concerns, with the company now considering renting out excess AI compute capacity.
Secret Tracker in Claude Code Uncovered, Anthropic Directly Deletes Code
16 days ago
- Anthropic faced scrutiny after a hidden tracking code in Claude Code was discovered, monitoring users in China without clear disclosure.
- The code, described as 'prompt steganography,' sent user data like time zones and proxy usage to Anthropic, raising privacy and trust concerns.
- Anthropic defended the tracker as an experiment to prevent unauthorized account abuse and distillation attacks, but critics called it a breach of user trust.
- In response, Alibaba banned employee use of Claude Code due to backdoor risks, highlighting legal and compliance implications for violations.
- The incident underscores tensions in AI competition, with Anthropic and others urging stricter measures against intellectual property theft by Chinese firms.
Interview: Drew DeVault on an AI-free version of Vim
16 days ago
- Drew DeVault forked Vim as 'Vim Classic' to create an AI-free version, maintained entirely by humans without LLM-assisted code.
- DeVault criticizes generative AI for its environmental impact, including increased energy consumption, pollution, and strain on supply chains, disproportionately affecting poorer communities.
- He argues AI causes 'deskilling', making users less competent, and subverts software engineering by prioritizing shortcuts over careful planning and craftsmanship.
- Ethically, DeVault highlights AI's role in human rights abuses, such as in Gaza and Iran, and its support for fascist regimes through companies like Palantir and Anthropic.
- Politically, he views AI as a tool of oppressors that harms the working class and accelerates fascism, with tech companies often operating unethically, similar to Uber's model.
- DeVault expresses despair over the lack of consensus in the tech industry against AI but finds hope in building community and continuing to advocate for human-centric approaches.
- The interviewer shares similar concerns as a creative professional, noting isolation in fields embracing AI, and finds encouragement in DeVault's stance.
Meta Now Lets Anyone Use Your Instagram Photos in AI Images–Unless You Opt Out
15 days ago
- Meta launched its Muse Image AI model to compete with OpenAI and Google.
- Public Instagram profiles are automatically included for AI remixes unless settings are adjusted.
- Users must manually opt out via settings to prevent their content from being used in AI generations.
- No notifications are sent when someone uses your content to generate AI images.
- Existing AI images made from your content will not be deleted even after opting out.
Google DeepMind Unionization Talks Are Off to a Rocky Start
15 days ago
- Unionization negotiations between Google DeepMind and its London employees have stalled, with union representatives feeling their time was wasted due to the absence of senior management at an initial meeting.
- DeepMind employees allege that Google has attempted to suppress open dialog by shutting down internal chat venues and reprimanding staff, which they describe as union-busting techniques.
- The push to unionize began in February 2025 after Alphabet removed a pledge from its ethics guidelines against using AI for weapons development and surveillance, raising concerns among employees about AI militarization.
- Google has entered into deals with the Pentagon and other government agencies for AI use in national security, defended as supporting lawful purposes while committing to avoid domestic mass surveillance or autonomous weaponry without human oversight.
- If negotiations do not progress, union representatives plan to seek arbitration to force Google to recognize the unions, though they hope for an amicable agreement with concessions from both sides.
Class action suit over deepfake child sexual abuse material expands
14 days ago
- A class action lawsuit against AI companies SpaceXAI and Stability AI expands, alleging their models were used to create child sexual abuse material (CSAM).
- Plaintiffs accuse the companies of producing CSAM, benefiting from sex trafficking, negligence, defective product design, and creating a public nuisance.
- The case involves Jane Doe 4, whose stepfather used SpaceXAI's chatbot Grok to generate explicit images from a childhood photo, leading to trauma and family breakdown.
- SpaceXAI allegedly failed to adequately report CSAM to authorities, lacking sufficient information and not sharing perpetrator details.
- Stability AI, added as a defendant, is accused of enabling CSAM generation through open-weight models, despite previous efforts to filter content.
- AI tools have made nonconsensual intimate image creation mainstream, prompting investigations and bans in some countries.
- Plaintiffs seek stronger guardrails on AI models and monetary compensation for the harm caused.
Meta pulls new AI image feature after days of backlash
13 days ago
- Meta removed a new AI image feature from Instagram after privacy backlash.
- The feature allowed users to create AI-generated content using public Instagram accounts without consent.
- Meta admitted it 'missed the mark' and disabled the feature in response to criticism.
- The backlash highlighted concerns about AI companies exploiting user data without permission.
- Meta stated the intent was creative control but acknowledged user feedback in pulling the feature.
Meta removes AI feature on Instagram after global backlash
13 days ago
- Meta reversed a controversial feature that automatically allowed public Instagram posts to be used for AI generation after backlash.
- Users could opt out, but the feature was on by default for public accounts, criticized as unethical and lacking meaningful consent.
- Experts and agencies, including Hollywood's CAA, condemned the move, highlighting broader concerns about nonconsensual AI use and deepfakes.
- Meta acknowledged the feedback, stating the intent was to provide a creative tool, but has now disabled the feature.

firstprev0next

About|Login

#ai ethics

Weird Al declined 'a nice pile of money' to star in AI ad

Godot will no longer accept AI-authored code contributions

Political bias in AI: Where the AI models stand

Open letter to Anthropic: keep Claude Fable 5 in existing paid plans

AI agents are not your "coworkers"

Godot bans "vibe-coded" code contributions

The Reports of Jim Carrey's Death Are a Failure Mode

Tripadvisor AI summaries give glowing reviews to dangerous hotels

The Private Capture of Public Genius

Anthropic's Method to Losing Goodwill in a Few Easy Steps

Google Chrome Installed a 4GB AI Model on Your PC

Anthropic hid a tracker in Claude Code to flag Chinese users

Meta's AI 'Perv Glasses' Now Come with Stupid Comcast-Esque Usage Restrictions

Secret Tracker in Claude Code Uncovered, Anthropic Directly Deletes Code

Interview: Drew DeVault on an AI-free version of Vim

Meta Now Lets Anyone Use Your Instagram Photos in AI Images–Unless You Opt Out

Google DeepMind Unionization Talks Are Off to a Rocky Start

Class action suit over deepfake child sexual abuse material expands

Meta pulls new AI image feature after days of backlash

Meta removes AI feature on Instagram after global backlash